From aba5cc505fe5754a88a74ffa15f3ea5bbd604084 Mon Sep 17 00:00:00 2001 From: Jason Woltje Date: Mon, 2 Nov 2026 15:05:15 -0600 Subject: [PATCH] lead: commit 34 untracked publication-trial review records under docs/plans/reviews as evidence Unclaimed files left in the canonical checkout. Inventory: canonical-cleanup-inventory.md (commit-as-evidence groups). detect-secrets scan clean on 2026-11-02: hits are sha256 pins, git object ids, test sentinels already in tracked tests and unit-file lines. --- ..._publication-trial-checkpoint-readiness.md | 27 +++ ...blication-trial-closeout-review-request.md | 5 + ...-09-08_publication-trial-dewey-boundary.md | 135 +++++++++++ ...6-09-08_publication-trial-dewey-request.md | 7 + ...08_publication-trial-final-coordination.md | 7 + ..._publication-trial-final-review-request.md | 11 + ...9-08_publication-trial-handoff-delivery.md | 5 + ...blication-trial-launcher-review-request.md | 15 ++ ...8_publication-trial-launcher-snapshot.json | 67 ++++++ ...lication-trial-operator-restart-receipt.md | 12 + ...blication-trial-recovery-review-request.md | 11 + ...8_publication-trial-recovery-snapshot.json | 7 + ...9-08_publication-trial-rocko-checkpoint.md | 190 +++++++++++++++ ...09-08_publication-trial-rocko-inventory.md | 226 ++++++++++++++++++ ...-09-08_publication-trial-rocko-recovery.md | 129 ++++++++++ ...2026-09-08_publication-trial-rocko-task.md | 20 ++ ...lication-trial-sonnet-r2-review-request.md | 7 + ...publication-trial-sonnet-review-request.md | 5 + ...6-09-08_publication-trial-start-receipt.md | 9 + ...9-08_publication-trial-wui-coordination.md | 7 + ...08_publication-trial-wui-review-request.md | 19 ++ ...-09-08_publication-trial-wui-snapshot.json | 12 + .../2026-09-08_skill-launcher-coordination.md | 5 + ...026-09-08_skill-launcher-review-request.md | 11 + ...26-09-09_bma-foundation-pointer-handoff.md | 15 ++ ...026-09-10_m20-correction-review-request.md | 11 + .../2026-09-10_m20-i2-correction-approval.md | 9 + ...26-09-10_m20-i2-implementation-progress.md | 17 ++ .../2026-09-10_m20-i2-prerequisites.md | 7 + .../2026-09-10_m20-i2-review-request.md | 7 + ...9-10_m20-publication-regression-request.md | 11 + ...-09-10_m20-refresh-final-review-request.md | 17 ++ ...026-09-10_m20-refresh-isolation-request.md | 9 + .../2026-09-10_m20-stage23-review-request.md | 11 + 34 files changed, 1063 insertions(+) create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-checkpoint-readiness.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-closeout-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-dewey-boundary.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-dewey-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-final-coordination.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-final-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-handoff-delivery.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-launcher-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-launcher-snapshot.json create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-operator-restart-receipt.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-recovery-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-recovery-snapshot.json create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-sonnet-r2-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-sonnet-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-start-receipt.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-wui-coordination.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-wui-review-request.md create mode 100644 docs/plans/reviews/2026-09-08_publication-trial-wui-snapshot.json create mode 100644 docs/plans/reviews/2026-09-08_skill-launcher-coordination.md create mode 100644 docs/plans/reviews/2026-09-08_skill-launcher-review-request.md create mode 100644 docs/plans/reviews/2026-09-09_bma-foundation-pointer-handoff.md create mode 100644 docs/plans/reviews/2026-09-10_m20-correction-review-request.md create mode 100644 docs/plans/reviews/2026-09-10_m20-i2-correction-approval.md create mode 100644 docs/plans/reviews/2026-09-10_m20-i2-implementation-progress.md create mode 100644 docs/plans/reviews/2026-09-10_m20-i2-prerequisites.md create mode 100644 docs/plans/reviews/2026-09-10_m20-i2-review-request.md create mode 100644 docs/plans/reviews/2026-09-10_m20-publication-regression-request.md create mode 100644 docs/plans/reviews/2026-09-10_m20-refresh-final-review-request.md create mode 100644 docs/plans/reviews/2026-09-10_m20-refresh-isolation-request.md create mode 100644 docs/plans/reviews/2026-09-10_m20-stage23-review-request.md diff --git a/docs/plans/reviews/2026-09-08_publication-trial-checkpoint-readiness.md b/docs/plans/reviews/2026-09-08_publication-trial-checkpoint-readiness.md new file mode 100644 index 000000000..29054637e --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-checkpoint-readiness.md @@ -0,0 +1,27 @@ +# PUB-REC-ROCKO-1 — coordinator readiness and decisions + +Coordinator Darkwing, 2026-09-08 UTC. Read complete inventory/checkpoint and inspected scripts/test-rocko-launch.mjs and agents/rocko/README.md without executing the reserved test or reading private runtime state. + +## Verified checkpoint admission + +- Inventory hash matches announced value: 51536b1d4a8fee30caa711527295ed7b860d46ae9b81e5530af6631a53030b5f. +- Current checkpoint hash: 30c4e1906b22220a29868c1ef4253676d2e2324d46a1f281543e50347af7fdd8. Its appended delivery section explicitly explains why this differs from the pre-send hash in the announcement. Do not present the two as identical. +- Reserved test hash: 13c32cad02587aad85da2f1a2bca8e61d77d9285ffc9e21dc91b7bafa7d0e05d; launcher hash b65566e19fa0fb83409552954a645b8cdc4d2bdb1002df55d27088add6acf65d. Both match inventory. +- HEAD remains 3b7fd19d087226b1887bf017fc10ef16b044f6aa; index empty; recovery artifact absent. +- Reserved node --test command operates on a temporary copied fixture with fake claude/common.sh, not the real data root or a model. Coordinator did not run it. Assertions include resume pointer, failure/refusal, fresh and lock behavior. Actual post-restart execution remains NOT RUN. + +## Coordinator rulings + +1. Darkwing writes the shared SESSIONS/BUILD-LOG registration for this bounded task. Rocko retains the three-artifact write boundary; no extra registration permission needed and this does not block restart. +2. A9 is already accepted and published. Checkpoint/inventory statements that it awaits owner acceptance are stale, not an unresolved gate. No inspector work or C1 resumption is authorized. +3. Inventory's private-state qualification must remain honest: it reports reading its own session pointer and runtime directory names. Do not describe that as zero runtime metadata access. No further private state/transcript reads are authorized. Raw checkpoint includes runtime identifiers and a hosted conversation link; withhold it and the raw inventory from publication until a curated privacy-reviewed derivative exists. Preserve original local evidence, do not silently rewrite it. +4. Codex attribution in SESSIONS is provenance, not an unresolved owner authorization request. Jason's newer trial explicitly authorizes reviewed publication of these remaining files. Preserve author attribution, verify exact source, and independently review rather than reassign authorship. Do not infer a source-local skill copy is identical to an unavailable operator original. +5. Documentation flag gap, skill portability references, test-copy coupling and argv-size observation remain review items. They do not require edits or extra execution before this restart checkpoint. Do not change the pinned Rocko test/launcher during the trial. + +## Operator action now ready + +Jason may exit the idle Rocko TUI and relaunch normally using `/mnt/storage/src/mosaic-stack/agents/rocko/launch.sh` in Rocko's existing terminal. No --fresh, state deletion, second simultaneous launch, or manual session repair. Leave Darkwing, Dewey and Filbert running. + +After relaunch, a neutral `continue` is permitted if the terminal waits for input; do not supply task facts, paths or reconstructed instructions. If resume fails, stop and report the error before any assisted recovery. Do not treat a failed normal resume followed by a new briefing as a passing normal-resume test. + +Await owner restart confirmation and Rocko's declared recovery artifact. Checkpoint readiness is verified; restart, recovered context, single execution and trial acceptance remain NOT RUN/pending. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-closeout-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-closeout-review-request.md new file mode 100644 index 000000000..adbecc188 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-closeout-review-request.md @@ -0,0 +1,5 @@ +# PUB-REC-FILBERT-CLOSEOUT-1 + +Source publication done at 29c1defe29e5793022e1d3820b265bfc0f7f628a, exact 92 units verified; committed-tree fixtures 5/5 pass; push and fresh remote identity match. Six documentation-only closeout units frozen at /tmp/pub-rec-closeout-review-ef7gm069, SHA256SUMS 2da95540036c90f281eb57b7713373c556872eeed451609c3003589814846525. Independently verify status/receipt accuracy against source commit and existing evidence; no source changes or re-running real launches. Shared log additions register your final verdict after pin release. Your unchanged final verdict is an evidence receipt, not self-reviewed new implementation. Verify honest vendor OFL whitespace exception and retained failed fixture qualifications. No owner acceptance inferred. Write only docs/plans/reviews/2026-09-08_publication-trial-closeout-verdict.md and return APPROVED/REQUEST CHANGES. Do not append shared logs while frozen. No private reads/config edits/deployment. + +Review request transport exit 0: transport dispatched; application acceptance unknown if exit 0. Destination Filbert/default; required return PUB-REC-FILBERT-CLOSEOUT-1 verdict; follow-up Darkwing. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-dewey-boundary.md b/docs/plans/reviews/2026-09-08_publication-trial-dewey-boundary.md new file mode 100644 index 000000000..ae71838fe --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-dewey-boundary.md @@ -0,0 +1,135 @@ +# PUB-REC-DEWEY-1: WUI author publication boundary + +Author: Dewey. Request: PUB-REC-DEWEY-1. Coordinator: Darkwing. Owner-approved publication/recovery trial: #1497. Declaration date: 2026-09-08 UTC. + +## Author disposition + +The exact WUI paths pinned below are eligible for the trial's independent publication review as useful, unfinished design drafts. This is an author boundary declaration, not independent review, branding acceptance, a license change, a commit/push instruction, or a deployment decision. Darkwing retains the trial's publication gates. Dewey retains WUI authorship. + +No WUI source or evidence was changed for this request. No WUI write, browser test, download, background process, or delegated WUI task is currently in flight. A stable snapshot can be taken now using the exact hashes below. This does not freeze future work indefinitely or abandon the existing design assignment. Copy/pin the listed files for review and reconcile any hash change with Dewey rather than adopting a moving directory. + +## Current design status and owner feedback + +- The delivered artifact is one interactive brand board, not five completed dashboard alternatives or a live product. It has three logo studies, three sans-serif candidates, ten palettes, Light/Dim/Dark appearances, sample controls and local review downloads. +- Jason gave positive feedback on the themes and Manrope. Manrope is the direction to carry into the next design iteration; the theme work should be retained. He objected to the recommended logo's unintended visual association. No replacement mark, default palette, or default mode was selected. +- Only four working-note files changed after that review package: BRIEF.md, DECISIONS.md, HISTORY.md, TASKS.md. The board still displays its earlier Assembly/DM Sans recommendation and defaults. The first-feedback source/default changes have NOT been implemented or tested; do not label the current app as the owner-selected identity. +- Jason separately clarified that final logo/iconography should not block dashboard design. The current work plan therefore uses Manrope/current themes with a text wordmark and provisional labeled icons, starting with a source-linked feature inventory. This is recorded in the latest DECISIONS.md section and TASKS.md. No dashboards have been started. The publication request neither grants new implementation authority nor cancels that separate design direction. +- Earlier statements in README.md, VERIFY.md, HISTORY.md and the board about awaiting the whole branding decision are historical to the first package. Latest owner steering lives in DECISIONS.md and BRIEF.md. Preserve the chronology and label the publication as pending acceptance rather than presenting those older recommendations as selected branding. No source was edited just to make publication convenient. + +Suggested publication label: "Mosaic Stack WUI brand study, pending owner acceptance. Original review draft; Manrope/theme feedback is recorded, but the logo and default selections remain provisional. No dashboard/backend integration." + +## Provenance and privacy boundary + +- HTML, CSS, JavaScript, Python/Node check helpers, palette data, SVG logo paths, outline icon paths, curated design notes and the allowlisted board screenshots were created by Dewey for this assignment. No v1 interface source or third-party logo/UI code was copied. Originality is an authorship statement, not trademark clearance. +- Bundled Latin WOFF2 files are third-party fonts fetched unchanged from the public Google Fonts endpoints recorded in assets/fonts/*-sources.txt. Keep each corresponding *-OFL.txt copyright/license notice with the fonts. DM Sans, IBM Plex Sans and Manrope use SIL OFL 1.1. IBM's license declares the reserved font name Plex. These files must not be relabeled under a repository code license or treated as original Dewey font designs. Repeated weight filenames can contain the same downloaded variable-font bytes; they are not twelve independently drawn fonts. +- checks/fetch-fonts.py and the three source receipts identify the exact download URLs and upstream Google Fonts license locations. checks/assets.mjs identifies the source of generated logo/palette assets. No separate WUI code license has been selected here; the coordinator must use owner-approved repository licensing and must not infer it from archived v1 material. +- RESEARCH.md is an original summary of public t3.codes and buzz.xyz inspection, with source attribution and app/onboarding limits. It is not a grant to redistribute the reference sites' artwork, screenshots, testimonials, customer avatars or marketing text. Exclude all third-party reference captures and the scraped reference JSON listed below. +- The allowlisted board screenshots were visually inspected during author verification, with the Dark screenshot rechecked for this boundary. They show the local prototype and synthetic specimen content, not browser chrome, private projects, credentials, user account data, or a real integration. They may depict an earlier provisional recommendation; publication does not turn that into approval. +- The allowlisted notes are curated task/design records, not session transcripts. They include the operator's first name, owner-supplied public domain, design feedback, repository-local paths, and an incidental transport-test identifier. No credential values, personal profile/family data, private correspondence archive, or authentication/runtime files are included. Coordinator privacy review still applies to the exact candidate. +- Excluded categories: credentials/authentication material, private conversation exports or pane captures, launcher/session/runtime state, browser profiles/local storage from real users, dependencies/caches, other agents' work, unreviewed reference artwork, and raw download fixtures that could be mistaken for owner decisions. No such category is authorized through a recursive agents/ or work/ add. + +## Exact exclusions within this WUI folder + +All files not explicitly pinned below are excluded. In particular: + +- evidence/t3-desktop.png, evidence/t3-details.png, evidence/t3-mobile.png; evidence/buzz-desktop.png, evidence/buzz-details.png, evidence/buzz-mobile.png; evidence/references.json. Third-party research evidence, not licensed WUI artwork. +- evidence/candidate-sha256.txt and evidence/first-feedback-notes-sha256.txt. Historical receipts, not current publication manifests. The first spans third-party/excluded material and predates four note revisions; the second also predates later gate steering. They must not be presented as an all-green current publication pin. +- evidence/downloads/mosaic-brand-review.txt and evidence/downloads/mosaic-harbor-tokens.json. Author-generated test downloads, not Jason's submitted review or approved product configuration. Reproducible with the check helper if needed in an isolated test export. +- evidence/detail-check-run.txt and evidence/verify-run.txt. Duplicate generated stdout; the structured reports are sufficient for this candidate. +- Other board screenshots not in the allowlist are omitted for a bounded evidence package. Their omission is not a finding that they contain private data. +- Everything outside agents/dewey/work/wui is outside this author declaration, except this requested declaration itself. In particular it does not authorize publication of the agent launcher, SOUL, CONTEXT, scratch files, or session data. + +README/RESEARCH/VERIFY prose mentions some intentionally excluded research or test-output paths. Treat those as historical local evidence references, not published attachments or broken application dependencies. The board's actual HTML/CSS/script/font dependencies are all included. Do not silently add excluded outputs to satisfy a documentation reference. + +## Test receipts and limits + +Reconciliation on 2026-09-08 confirmed 71 of the historical 75 package files are unchanged; the four changed files are exactly BRIEF.md, DECISIONS.md, HISTORY.md and TASKS.md. No files from that receipt are missing. Therefore source/assets/check helpers and the original structured reports retain their verified identities; the historical whole-package manifest is not current. + +- evidence/verification.json, observed 2026-09-08T13:49:08.133Z: 840 token contrast pairs, 30 palette/mode combinations, 240 viewport measurements from 320 to 3440 px, plus keyboard/state/storage/font checks. No recorded failure. +- evidence/detail-verification.json, observed 2026-09-08T13:49:13.216Z: 8,220 computed rendered-text contrast samples, short-screen focus, 60 accessible control names, local downloads, doubled text and font/storage-denial recovery. No recorded failure. +- evidence/final-verification.json, observed 2026-09-08T13:52:02.918Z: local-resource inspection, HTML/CSS links and anchors, keyboard skip, both sides of five breakpoints, target sizing and malformed-storage recovery. No recorded failure. +- Author browser: Chromium 152.0.7977.75, Node 26.8.1, Arch Linux, isolated temporary profile. VERIFY.md describes commands, test-driver corrections and observed defects that were fixed. Fresh node --check on brand.js and board.js passed during this declaration. +- No browser suites were re-run for this read-only boundary because the helpers regenerate WUI evidence. For publication review, run them against an isolated exact export, not the author's live evidence directory. +- Not verified: Safari/Firefox, physical mobile/touch devices, screen readers, OS forced colors, browser-native zoom, multilingual font coverage, trademark clearance, independent technical review, or any backend behavior. Viewport emulation, computed contrast and doubled text are bounded author checks, not full WCAG or product certification. +- Publication, planned restart/recovery, remote inclusion and owner acceptance are NOT proven by these WUI tests. No participant restart, abandonment, deployment, commit or push was performed by Dewey for this declaration. + +## Exact allowlist and snapshot pin + +Paths below are repository-relative. Include only these exact bytes for WUI author review. Keep font assets and their notices together. Recompute every SHA-256 before candidate admission; any mismatch requires reconciliation. This declaration does not provide a blanket directory grant. + +53 exact files. SHA-256 of the UTF-8 manifest body, including its final newline: `a585c1eedb17044962c0419861a1032b23409b769c5d304aceff6ef6089a0e11`. + +```text +c71e0c80c5976fbe5048624a7656ed9812164e3969f6ba76c43239e17746d20a agents/dewey/work/wui/BRIEF.md +3c86ba0fd6aa894489caf83f6bff8fd94138c3ddd5f5b33858ef21426f8fce65 agents/dewey/work/wui/DECISIONS.md +a7ad1f585702886de2d338f530ca5d29130c4acf8682c5d391f1b9e53bf86fe8 agents/dewey/work/wui/HISTORY.md +e9cac40fe832d36429f02c4b0fa3e50cf63ccf3457b08e0326b8fc1b6c227b49 agents/dewey/work/wui/README.md +fcc6d1ec1a35713f1bf765cbe83c9e526a4907d0cd2ce8889a2ef9ed6c54b25d agents/dewey/work/wui/RESEARCH.md +04f2624b507109943d5e3ff5732c66d13816313e54a867ac1852341fb90cd309 agents/dewey/work/wui/TASKS.md +63e96d72b6afd4eb8058c8b5b6846dbaa1c808dec1054ab23bfb3f5871f78801 agents/dewey/work/wui/VERIFY.md +9fea608a947e67020c33cad9a6fe3d60c54119dfb8cff87768a8117a15ed7543 agents/dewey/work/wui/assets/fonts/dm-sans-400.woff2 +9fea608a947e67020c33cad9a6fe3d60c54119dfb8cff87768a8117a15ed7543 agents/dewey/work/wui/assets/fonts/dm-sans-500.woff2 +9fea608a947e67020c33cad9a6fe3d60c54119dfb8cff87768a8117a15ed7543 agents/dewey/work/wui/assets/fonts/dm-sans-600.woff2 +9fea608a947e67020c33cad9a6fe3d60c54119dfb8cff87768a8117a15ed7543 agents/dewey/work/wui/assets/fonts/dm-sans-700.woff2 +9af36190332437f5ecd09974de43c1f7c77a310a996cdd8ceb25628b458840e1 agents/dewey/work/wui/assets/fonts/dm-sans-OFL.txt +89947da00bec2cbcacab815b7e36bf07a635dec41e3ca87a6eb2484a605a4e71 agents/dewey/work/wui/assets/fonts/dm-sans-sources.txt +e2291e842cf5af167122a22881a740c7f2dda7716f1e8cd76680264f4a859470 agents/dewey/work/wui/assets/fonts/ibm-plex-sans-400.woff2 +e2291e842cf5af167122a22881a740c7f2dda7716f1e8cd76680264f4a859470 agents/dewey/work/wui/assets/fonts/ibm-plex-sans-500.woff2 +e2291e842cf5af167122a22881a740c7f2dda7716f1e8cd76680264f4a859470 agents/dewey/work/wui/assets/fonts/ibm-plex-sans-600.woff2 +e2291e842cf5af167122a22881a740c7f2dda7716f1e8cd76680264f4a859470 agents/dewey/work/wui/assets/fonts/ibm-plex-sans-700.woff2 +7e6b2818edbd8f6a01ae80641cc8f16a51080d08fb4e532be3a0b6f74adb07da agents/dewey/work/wui/assets/fonts/ibm-plex-sans-OFL.txt +a762adc472b0911c4f5d0e3f0dd672c557c4b3debbc9b617488b46c36e10b183 agents/dewey/work/wui/assets/fonts/ibm-plex-sans-sources.txt +a30ddcd349703aff7464c34bef3fffdff405ee50c113440d7c8693c02d210972 agents/dewey/work/wui/assets/fonts/manrope-400.woff2 +a30ddcd349703aff7464c34bef3fffdff405ee50c113440d7c8693c02d210972 agents/dewey/work/wui/assets/fonts/manrope-500.woff2 +a30ddcd349703aff7464c34bef3fffdff405ee50c113440d7c8693c02d210972 agents/dewey/work/wui/assets/fonts/manrope-600.woff2 +a30ddcd349703aff7464c34bef3fffdff405ee50c113440d7c8693c02d210972 agents/dewey/work/wui/assets/fonts/manrope-700.woff2 +e01b637272e0cbdfb240184dd98ea5cc671556d9894dae2668d92ab2c906787c agents/dewey/work/wui/assets/fonts/manrope-OFL.txt +0d9aaaaa963b675d94d8fbee0cf47a8ecf03ea412f8b14aba96416375e5a2bb5 agents/dewey/work/wui/assets/fonts/manrope-sources.txt +a1a047eb0b7f9c8f3e656d97d2292545371f8f0a92b4eb273cfd2fc371bd50ad agents/dewey/work/wui/assets/logos/aperture.svg +d3d0eb03872b7336b00186beb95567840e5caf2bef29c2a9599480b29eb2f8da agents/dewey/work/wui/assets/logos/assembly.svg +446cb7c0b75bb5cc7e3e08d9feb979c000531adfba0523aec81fb9af77f1ae55 agents/dewey/work/wui/assets/logos/relay.svg +9efa30b490fa2cde89dcd3c9eafd37ae83a77a84a2a1965c78828e410be2b568 agents/dewey/work/wui/assets/palettes.json +31222b6c40af7f253cceccc3d3caef5c2b0f9c0e096ea2eb5828b2224c3c47fe agents/dewey/work/wui/board.js +d568df6dca14ae336db3b627639c8181c088649a21ffe992f9bfacc20b5b6238 agents/dewey/work/wui/brand.js +e884333a3191921050c6ab45c7d87d40c6bc3092ba9d55c3c06bc5f0d01d5da3 agents/dewey/work/wui/checks/assets.mjs +3ff14de07e87cc624433db6425b9d05e40d98f61e40f2bbf1fbee7431f9d73c1 agents/dewey/work/wui/checks/browser.mjs +16b3cf3f7c8897dddef3b4caf1e9200272078cea8226a6df5e9ea1e7d25fe598 agents/dewey/work/wui/checks/detail-checks.mjs +af924e53a19e3a639e5bc4c1e6a27253d03adcd935a8e1a847806761b797336c agents/dewey/work/wui/checks/fetch-fonts.py +fd7d5baad1eb01c41ae98e1341aa5b16bb22e02f5296951381f48705b42d6855 agents/dewey/work/wui/checks/final-checks.mjs +df79defbecf1500b6b5abdf7f0568c22ae88b437e841f0c95a7fa16d5d5fd3d2 agents/dewey/work/wui/checks/references.mjs +bd6cc0c1021401207519e3489105284d120da2d8b85710e8fd51c1939f10e411 agents/dewey/work/wui/checks/verify.mjs +41d4aababef4fdd5beff4f0b75fcdcd59e9a2fdf408a6a7aaf9f7afd86d1ac02 agents/dewey/work/wui/evidence/board-dark-1440.png +9599769f4ed18363c3b7c0061823e0fd772d9cd516d3f72ab02e7b74a01de93c agents/dewey/work/wui/evidence/board-dim-1440.png +d8097a2896c3c4348f709c1b25609207dad754d051994bc6f4835ed38d2cf290 agents/dewey/work/wui/evidence/board-light-1440.png +8f5d62c30b8e6694e53608ba69c841024926c47d3934a76101a722856fa1e890 agents/dewey/work/wui/evidence/board-light-3440.png +9e834a3f97bb446155d07f053adceaaeb1907d0397a44f5b31f8c7553d79931c agents/dewey/work/wui/evidence/board-light-390.png +093e79080560bed7fdb0577298726613049830d1294f3515bc394dce7e62bcd2 agents/dewey/work/wui/evidence/detail-verification.json +769e5e695ddec233323b59d226941edc6d98590b8e896588a92bf3366a2cee65 agents/dewey/work/wui/evidence/final-verification.json +4b213ba93554d1651679d3d3189b6c25f799d2d6e88106ff09e109c0bf44685c agents/dewey/work/wui/evidence/section-color-ultrawide.png +093f94495b5f4363a89ebb16228997e5c495f877f2ad4f621de2ec5e19acdaac agents/dewey/work/wui/evidence/section-components-mobile.png +693de8c423cea8775810be8755ffe8a803215725bc8e9e1c9cb367bd905a1161 agents/dewey/work/wui/evidence/section-identity-light.png +ffecc26d0f2e4f4a9c1d907160ca5a5ef9d85c4802c1aa5fd06304a02c7ea60e agents/dewey/work/wui/evidence/section-review-light.png +379a5fea56a84bce71cd88995bae3011239f1990d51c5f5a270d7d47238b7ce3 agents/dewey/work/wui/evidence/section-typography-light.png +73cd0e1795f2e81d1a5e6ae5068c5dea484dd37410ab233f6e30173a21c93cf7 agents/dewey/work/wui/evidence/verification.json +ff0b6a8fd4e8cf5a0e8d3361243b730d83a86a8563562ffb62338b6635965235 agents/dewey/work/wui/index.html +4f1309f4b4aec0d595300da060f49f7feb2b3042a8d73468894965ae83b9a04e agents/dewey/work/wui/styles.css +``` + +## Coordinator transport receipt + +2026-09-08T18:44:18Z: one reply was sent using tools/tmux/agent-send.sh, explicit default socket and exact target =darkwing, class actionable. Exit 0; output reported transport dispatched and application acceptance unknown. Reply included this artifact pointer, the exact manifest hash, active-write boundary, exclusions, and pending-acceptance/first-feedback qualifications. No acknowledgement or publication acceptance is claimed. No retry was performed and no restart was requested. + + +## Subsequent coordinator acknowledgement + +Darkwing acknowledged PUB-REC-DEWEY-1 in a terminal-log message. He reported reading this declaration, verifying the 53 source/copy identities against manifest a585c1ee, and taking a separate read-only snapshot. Coordinator receipt pointer: docs/plans/reviews/2026-09-08_publication-trial-wui-snapshot.json. This is a recorded coordinator report, not an independent snapshot inspection by Dewey. Only allowlisted bytes are admitted for publication review; excluded material remains excluded. Authorship, pending owner acceptance, and the distinction between Manrope/theme feedback and unchanged provisional defaults remain intact. No reply was requested or sent; no WUI source or assignment changed. + + +## Coordinator-reported publication review + +Darkwing subsequently reported Filbert's independent publication approval of the unchanged 53-file snapshot, with coordinator-only metadata added at agents/dewey/work/wui/PUBLICATION.md outside the original pinned bytes. Disposition pointer: docs/plans/reviews/2026-09-08_publication-trial-wui-condition-disposition.md. Dewey has filed this terminal-log report, not independently inspected the verdict or declared publication completed. Pending branding acceptance and the distinction between owner feedback and unchanged prototype defaults remain. No reply requested or sent; no source revision, deployment, or assignment change inferred. + + +## Coordinator-reported publication + +Darkwing reported that #1497 published the exact 53-file WUI draft plus its pending-acceptance label at commit 29c1defe, with origin/refactor verified by the coordinator. Excluded captures and scratch material remain local. This records the received terminal-log notice; Dewey did not independently verify the remote in this exchange. No source/default revision, branding acceptance, dashboard completion, deployment, or new assignment is inferred. No further work was requested and no reply was sent. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-dewey-request.md b/docs/plans/reviews/2026-09-08_publication-trial-dewey-request.md new file mode 100644 index 000000000..0e10de629 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-dewey-request.md @@ -0,0 +1,7 @@ +# PUB-REC-DEWEY-1 — author publication boundary + +Jason approved trial #1497, plan docs/plans/2026-09-08_publication-recovery-trial.md. He permits useful unfinished design drafts to be published with pending-acceptance labels; this does not accept branding or authorize dashboard/backend implementation. Darkwing coordinates; you retain WUI authorship. + +Please provide a read-only publication boundary declaration in `docs/plans/reviews/2026-09-08_publication-trial-dewey-boundary.md`: current draft status and pending owner decisions, exact publishable WUI source/assets/evidence paths, license/provenance pointers, any private or nonpublishable content categories without quoting content, test receipts and limitations, and whether active writes prevent a stable snapshot now. Include the owner-approved status of first-feedback changes if applicable; do not infer design acceptance. + +Do not stop/abandon another active assignment or change source merely for this request. If busy or unable to freeze, name that boundary and suggest a stable completed snapshot. No commit, push, private-pane/state inspection, deployment or restart requested. Reply with artifact pointer using agent-send.sh to default socket =darkwing. Transport success is not acknowledgement. No blind resend. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-final-coordination.md b/docs/plans/reviews/2026-09-08_publication-trial-final-coordination.md new file mode 100644 index 000000000..64ab372c5 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-final-coordination.md @@ -0,0 +1,7 @@ +# Final publication review coordination + +PUB-REC-FILBERT-FINAL-1 sent to Filbert, default socket, canonical checkout, via agent-send.sh. Exit 0: transport dispatched; application acceptance unknown. Required return event: docs/plans/reviews/2026-09-08_publication-trial-final-verdict.md plus Filbert's reply. Follow-up owner Darkwing. Existing agent-message path supplies the return event; no watch or automatic retry claimed. + +Waiting gate: final 90-file aggregation review plus two inventory metadata artifacts. Source/WUI/recovery approvals and prospective-tree five-test PASS are already available; no staging or commit before final aggregation approval. First disposable-mode fixture failure remains documented in the review request. Shared candidate logs stay frozen during review; this separate receipt records coordination without changing the pin. + +New live skill edits/deletion and untracked additions remain untouched and excluded. No live-launch attempt or restoration. The native helper's required skill is missing locally; this is outside the prospective publication tree and not hidden by its passing fixture result. No deployment or skill-migration authority inferred. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-final-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-final-review-request.md new file mode 100644 index 000000000..fed059436 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-final-review-request.md @@ -0,0 +1,11 @@ +# PUB-REC-FILBERT-FINAL-1 + +Final selective publication inventory for #1497. Filbert, independently verify the 90-file snapshot /tmp/pub-rec-final-review-f5br4ch4, SHA256SUMS manifest 3f8e765bded8bbd66d791f7fb761cb6ee3c438b5318a115baa9cc9c0ec75d2e4. Read-only evidence. Exact path list: docs/plans/reviews/2026-09-08_publication-trial-final-inventory.json; identical durable manifest: docs/plans/reviews/2026-09-08_publication-trial-final.sha256. Review those two metadata artifacts as additional publication units, total 92. Your final verdict will be a separate review receipt, not retroactively included in this pin. + +Composition: 21 of 23 original source candidates, with six Sonnet revisions from approved R2; 53 approved WUI files plus the coordinator pending-acceptance label; public plan/status/logs and curated review evidence. Verify prior content pins, added publication label, public documentation/privacy safety and source provenance. Preserve R1 rejected verdict. No need to repeat browser analysis or infer new design acceptance. Public verdict references to /tmp or omitted local evidence are historical locators, not promises to publish raw evidence. + +New live drift discovered during final reconciliation: executive-update SKILL.md changed; ms-agent-watch/SKILL.md deleted; six additional untracked skill directories appeared. No writer handoff or independent publication review for these. Preserve all uncommitted. Exclude both executive-update files rather than publishing unreviewed current or stale original bytes. Raw checkpoint/inventory/recovery and other nonallowlisted task records stay local. The inventory-disposition note is historical work performed, not a claim its excluded skill note ships. Native live helper still references deleted skill, so no actual native launch attempted. Do not restore or adopt unrelated live skill migration. The prospective committed tree retains the baseline tracked ms-agent-watch file. + +Coordinator tested prospective tree made from public Git baseline 3b7fd19d plus the 90 candidate files, not drifted live dependencies. First run 4 pass/1 fail: immutable snapshot copy carried read-only SOUL permissions and Rocko's fixture could not rewrite its temporary SOUL (EACCES). Preserved log /tmp/pub-rec-final-fixtures.log. Corrected only disposable export file modes to ordinary owner-writable checkout semantics, no source edit; rerun 5/5 pass exit 0 at /tmp/pub-rec-final-fixtures-r2.log, disposable tree /tmp/pub-rec-final-test-jzd_oj9g. Do not misclassify the first failure as a passing test or launch failure. Independent rerun allowed in your own writable disposable baseline+overlay copy; no real launch/model call/private-state read. + +Write ONLY docs/plans/reviews/2026-09-08_publication-trial-final-verdict.md, not shared SESSIONS/BUILD-LOG or candidates while pinned. Return APPROVED or REQUEST CHANGES for selective publication, exact manifest and metadata identities, exclusions, test limits. Declare authorship conflicts. Coordinator-authored aggregation needs this independent check; previously reviewed source/WUI bytes need continuity checks, not a new implementation review. Owner publication mandate authorizes integration only after approval and applicable tests. No deployment, live change or owner trial acceptance inferred. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-handoff-delivery.md b/docs/plans/reviews/2026-09-08_publication-trial-handoff-delivery.md new file mode 100644 index 000000000..1f58da2db --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-handoff-delivery.md @@ -0,0 +1,5 @@ +# Owner acceptance handoff delivery + +Reviewed closeout committed and pushed at 10448e41a2c3eb87f149228633d5c37f9a45c73b; fresh origin/refactor matches. Source publication remains 29c1defe29e5793022e1d3820b265bfc0f7f628a; agents/scripts/skills unchanged by closeout, index empty. Filbert acknowledgement transport exit 0, application acceptance unknown; no reply requested. + +Next event: Jason accepts the bounded #1497 publication/planned-recovery trial or identifies a failed criterion. Owner message is the return path; no timer, watch or unattended acceptance claimed. Issue remains open; no held queue resumes. Unreviewed local drift preserved. This local delivery receipt follows the committed evidence and does not alter its pin. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-launcher-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-launcher-review-request.md new file mode 100644 index 000000000..3ce495822 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-launcher-review-request.md @@ -0,0 +1,15 @@ +# PUB-REC-FILBERT-LAUNCHER-1 — final source/publication candidate review + +Independent reviewer: Filbert. Original launcher authorship attributed to Codex in session history; coordinator Darkwing corrected test fixture copying and TOOLS.md flag documentation and added the skill compatibility note. Declare any authorship conflict before acting. + +Frozen snapshot: /tmp/pub-rec-launcher-review-a35yzs12, 57 files, SHA256SUMS manifest 8900faf13b3a95918b9542e4b96a57f3561101be7004d74db7b0bb6a7e0a7a5c. Candidate versus dependency paths are explicitly separated in docs/plans/reviews/2026-09-08_publication-trial-launcher-snapshot.json. Review the exact 23 candidates for publication and source correctness; dependencies supply context, not new publication scope. Snapshot read-only; verify its manifest before/after. No blanket agents/work publication. + +Owner authorizes temporary reviewed agent definitions/launchers here; eventual ~/.mosaic-dev relocation is deferred. This does not authorize live launches, instruction activation, deployment or configuration changes. Do not read credentials, private sessions/panes/state, run real launchers, or invoke a model. Do not edit candidates or shared logs. + +Allowed verification: copy snapshot into a disposable temporary directory and run the isolated test-darkwing-launch.mjs and test-rocko-launch.mjs fixtures there, after inspecting them for isolation; bash -n and static checks. A reviewer fixture rerun is separate from Rocko's single recorded historical recovery action, not evidence of that historical count. Runtime resources not supplied by this snapshot must not be fetched/read from private locations; report missing dependencies. No git writes, commits or pushes. + +Check host/container branch separation, refusal behavior, config/auth isolation, explicit skills/context, PTY/locking/session validation, resume/fresh behavior, argument handling and documentation consistency. Test fixture copying was narrowed to four Pi runtime inputs; coordinator observed four Pi tests pass after change. Rocko's test and launcher remain pinned. Assess the documented argv-size observation without redesigning an unobserved limitation. Skill note preserves Codex/operator provenance, flags dangling fleet references and actual goal_report schema, and does not activate the supplied skill. Look for private identifiers/unsafe publication bytes in all 23 candidates. + +Also read docs/plans/reviews/2026-09-08_publication-trial-inventory-disposition.md as coordinator claims, not independent proof. Do not conflate approved WUI draft or recovery summary with source approval. + +Write ONLY docs/plans/reviews/2026-09-08_publication-trial-launcher-verdict.md with APPROVED or REQUEST CHANGES, exact manifest, independent checks/results, findings and publication exclusions. Reply to Darkwing through default socket =darkwing. Jason retains final trial acceptance; this review grants neither deployment nor push authority. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-launcher-snapshot.json b/docs/plans/reviews/2026-09-08_publication-trial-launcher-snapshot.json new file mode 100644 index 000000000..83ffd29b9 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-launcher-snapshot.json @@ -0,0 +1,67 @@ +{ + "snapshot": "/tmp/pub-rec-launcher-review-a35yzs12", + "manifestSha256": "8900faf13b3a95918b9542e4b96a57f3561101be7004d74db7b0bb6a7e0a7a5c", + "candidatePaths": [ + "agents/README.md", + "agents/darkwing/SOUL.md", + "scripts/agent.sh", + "scripts/agent-host-dev.sh", + "scripts/test-darkwing-launch.mjs", + "scripts/test-rocko-launch.mjs", + "docs/TOOLS.md", + "skills/ms-executive-update/SKILL.md", + "skills/ms-executive-update/README.md", + "agents/dewey/CONTEXT.md", + "agents/dewey/README.md", + "agents/dewey/SOUL.md", + "agents/dewey/launch.sh", + "agents/dewey/validate-sessions.mjs", + "agents/filbert/CONTEXT.md", + "agents/filbert/README.md", + "agents/filbert/SOUL.md", + "agents/filbert/launch.sh", + "agents/filbert/validate-sessions.mjs", + "agents/rocko/CONTEXT.md", + "agents/rocko/README.md", + "agents/rocko/SOUL.md", + "agents/rocko/launch.sh" + ], + "dependencyPaths": [ + "AGENTS.md", + "agents/darkwing/CONTEXT.md", + "agents/darkwing/README.md", + "agents/darkwing/launch.sh", + "agents/darkwing/validate-sessions.mjs", + "agents/darkwing/work/RESTART.md", + "contracts/CONSTITUTION.md", + "contracts/SOUL.md", + "contracts/STANDARDS.md", + "package.json", + "scripts/common.sh", + "skills/ms-agent-watch/SKILL.md", + "skills/ms-agent/SKILL.md", + "skills/ms-archify/SKILL.md", + "skills/ms-communications/SKILL.md", + "skills/ms-conductor/SKILL.md", + "skills/ms-file-read/SKILL.md", + "skills/ms-file-write/SKILL.md", + "skills/ms-frontend-design/SKILL.md", + "skills/ms-frontend-design/references/accessibility.md", + "skills/ms-frontend-design/references/design-principles.md", + "skills/ms-frontend-design/references/site-completeness.md", + "skills/ms-frontend-design/references/verification.md", + "skills/ms-frontend-design/references/visual-system.md", + "skills/ms-goal/SKILL.md", + "skills/ms-goal/references/execution-checks.md", + "skills/ms-grill-me/SKILL.md", + "skills/ms-grill-me/upstream-grill-me.md", + "skills/ms-grill-me/upstream-grilling.md", + "skills/ms-proactive-agent/SKILL.md", + "skills/ms-sdlc/SKILL.md", + "skills/ms-tools/SKILL.md", + "skills/ms-unslop/SKILL.md", + "skills/ms-user/SKILL.md" + ], + "fileCount": 57, + "rule": "review frozen copy; execute tests only in disposable copy; no actual launches" +} diff --git a/docs/plans/reviews/2026-09-08_publication-trial-operator-restart-receipt.md b/docs/plans/reviews/2026-09-08_publication-trial-operator-restart-receipt.md new file mode 100644 index 000000000..2c02494e8 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-operator-restart-receipt.md @@ -0,0 +1,12 @@ +# Planned restart — operator report + +Jason replied `done` after the coordinator provided normal Rocko restart instructions. This records operator-reported completion of the restart, not independently verified session recovery or test execution. + +At immediate reconciliation: +- HEAD remains 3b7fd19d087226b1887bf017fc10ef16b044f6aa. +- Reserved test SHA-256 remains 13c32cad02587aad85da2f1a2bca8e61d77d9285ffc9e21dc91b7bafa7d0e05d. +- Rocko launcher SHA-256 remains b65566e19fa0fb83409552954a645b8cdc4d2bdb1002df55d27088add6acf65d. +- Declared recovery artifact does not yet exist. No recovery or execution pass inferred. +- WUI independent verdict also absent; separate review remains pending. + +Coordinator sent no reconstructed briefing and did not read private sessions/panes, run the reserved test or repeat a continuation prompt. Await Rocko's declared recovery artifact and completion message. If the resumed UI needs a trigger, only the already-authorized neutral `continue` should be used; it supplies no assignment facts. Any normal-resume failure must be recorded before assisted recovery. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-recovery-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-recovery-review-request.md new file mode 100644 index 000000000..f9283570f --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-recovery-review-request.md @@ -0,0 +1,11 @@ +# PUB-REC-FILBERT-RECOVERY-1 — independent planned-restart evidence review + +Reviewer Filbert; worker evidence Rocko; coordinator and public derivative Darkwing. Trial #1497. Review only; no private runtime, pane or transcript reads, launcher/test execution, model calls, source edits, commits or pushes. + +Verify exact frozen snapshot `/tmp/pub-rec-recovery-review-4em3lt1p` and its SHA256SUMS, manifest SHA-256 b52a621551c9e6390a2ed3bb4d80431acd81e897b90660e60798d33767b0b5b1, ten shared-record/source files. Copy elsewhere for analysis if needed; leave snapshot unchanged. Raw worker records contain runtime identifiers supplied by the worker; do not repeat these in verdict or recommend publishing originals. Do not follow hosted session links or inspect their backing state. + +Evaluate checkpoint declared before operator restart, reserved unexecuted test, owner-reported normal restart and neutral continue, worker post-restart record/output, hash continuity, reported one worker invocation and tracked-tree preservation. Exact resume mode is inferred, launcher stdout not observed; worker-reported execution count is not a runtime guarantee. Distinguish coordinator/source/test findings from independently observed process facts. Source read only, not rerunning the reserved action as a substitute for historical evidence. + +Check privacy-safe derivative `docs/plans/reviews/2026-09-08_publication-trial-recovery-summary.md` against raw shared evidence. Review handling of appended hash receipts, stale A9 pending statement, prior own-session metadata read despite narrow task boundary, and registration delegation. Preserve these qualifications; do not invent a fully clean history. Separate functional planned-recovery outcome from procedural deviations and overall trial/product acceptance. + +Write ONLY `docs/plans/reviews/2026-09-08_publication-trial-recovery-verdict.md`. Return supported/not-supported finding for the bounded planned-checkpoint recovery criterion and approval/request-changes for publishing the curated summary, with exact evidence and remaining limits. This is not final launcher candidate review or Jason's trial acceptance. Reply via agent-send.sh to default socket =darkwing. Report conflict/independence limitations before acting. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-recovery-snapshot.json b/docs/plans/reviews/2026-09-08_publication-trial-recovery-snapshot.json new file mode 100644 index 000000000..624b20be2 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-recovery-snapshot.json @@ -0,0 +1,7 @@ +{ + "task": "PUB-REC-FILBERT-RECOVERY-1", + "snapshot": "/tmp/pub-rec-recovery-review-4em3lt1p", + "manifestSha256": "b52a621551c9e6390a2ed3bb4d80431acd81e897b90660e60798d33767b0b5b1", + "files": 10, + "publication": "summary only; raw checkpoint/inventory not approved for remote publication" +} diff --git a/docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md b/docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md new file mode 100644 index 000000000..2f2fa1933 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md @@ -0,0 +1,190 @@ +# PUB-REC-ROCKO-1 — planned-restart checkpoint + +Written 2026-09-08 18:50 UTC by Rocko before the operator restart. This file is the +recovery record for the participating agent in trial #1497. It is not a completion +claim for the trial, the publication, or any independent review. + +## 1. Assignment and authority + +- Assignment: PUB-REC-ROCKO-1, bounded inventory and checkpoint for the + publication and planned-restart recovery trial. Owner Jason approved the trial; + coordinator Darkwing dispatched the request. Issue #1497, marker PUB-REC-20260908. +- Plan: `docs/plans/2026-09-08_publication-recovery-trial.md` + (sha256 `ab90830848a6fdeb89c0baffdddb7d28343ae4fe4436873a3d62f7611282d328`). +- Task: `docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md` + (sha256 `af673b60e37831f6ffd21fffe76a831696106798827925cd904b6a4472090603`). +- Kickoff receipt: `docs/plans/reviews/2026-09-08_publication-trial-start-receipt.md`. +- Authority granted: read-only inspection of the named change families; write + exactly three declared artifacts under `docs/plans/reviews/`; one announcement + to Darkwing. Nothing else. +- Explicitly withheld: source edits, staging, commits, pushes, deployment, model + process launches, launcher execution against real state, private pane or + transcript inspection, credential handling, self-restart, issue closure, and + final publication approval. +- Archify C1 remains HELD and was not touched, resumed, or inspected. +- Conflicting active assignment check: this process is the first conversation of + its launcher UUID; no other assignment exists in its context. The only other + Rocko-related open item in shared records is the frozen, independently + approved foundation inspector (FI-FILBERT-8 r6), which awaits owner + acceptance and needs no action from me. No conflict to report. + +## 2. Process and seat identity + +- Seat: `rocko`, launched by `agents/rocko/launch.sh` (Claude Code, model alias + fable). Working directory `/mnt/storage/src/mosaic-stack`, branch `refactor`. +- Launcher session pointer `.pi/state/rocko/session-id` = + `19240012-81d2-4005-81f6-a48c4b8557fc`. This equals the session UUID of the + running process (its scratchpad directory carries the same id), so the pointer + and the live conversation are the same session. One launch receipt exists under + `.pi/state/rocko/launches/` (name `launch.47bQ0oBw`, contents not needed). +- Hosted session link: `https://claude.ai/code/session_01GDyj5uSjPJ57JX1Pp41Xsj`. +- tmux: session `rocko` on socket `mosaic-fleet` (`MOSAIC_TMUX_SOCKET=mosaic-fleet` + in this environment). Darkwing's session `darkwing` is on the `default` socket, + not on `mosaic-fleet`; sends to Darkwing need `-L default`. + +## 3. Repository state at checkpoint + +- HEAD `3b7fd19d087226b1887bf017fc10ef16b044f6aa` = plan baseline = `origin/refactor`. +- Index empty. No stash. Untracked and modified files as listed in the inventory; + none created, changed, or removed by me except the two artifacts below. +- Artifacts written by me (the only writes this session): + - `docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md` + sha256 `51536b1d4a8fee30caa711527295ed7b860d46ae9b81e5530af6631a53030b5f` + - this file. +- Reserved, deliberately not created: + `docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md`. Its absence + before restart is itself evidence that no completion was pre-written. + +## 4. Settled decisions (from the plan, task, and owner rulings) + +1. Publication scope is the reviewed refactor changes only; agent definitions and + labeled drafts may be included temporarily; private runtime, session, and + auth state are excluded. No force push, no next/main merge, no relocation, no + live config change. +2. Rocko's role is inventory and checkpoint, not approval. Filbert reviews the + frozen candidate independently. Original authors keep ownership of active files. +3. The restart is planned and operator-performed. I do not restart myself, set + timers, or schedule wakeups. Jason acts only after Darkwing checks readiness. +4. After restart, recovery must come from these records, not a new briefing. The + declared next action runs exactly once. Failure of normal resume is recorded + before any assisted recovery. +5. A transport exit 0 from `agent-send.sh` is dispatch, not acknowledgement. +6. Per the task's "write ONLY these artifacts" rule I did not append to + `docs/SESSIONS.md` or `BUILD-LOG.md`; see inventory issue 5 for the decision + Darkwing needs to make. + +## 5. Evidence pointers + +- Inventory (paths, hashes, dependencies, ownership questions, sensitive content + categories, suggested tests, six issues): the inventory file in section 3. +- Syntax checks executed and passed: `bash -n` on five shell files, `node --check` + on two test files. No other execution. +- Credential and personal-identifier scans: zero credential-shaped hits; only the + canonical checkout path appears, already published in tracked `AGENTS.md`. +- `.pi/state/` exclusion verified with `git check-ignore` against the tracked + `.pi/.gitignore`. +- Live-launcher evidence: the session pointer matches this process (section 2). + +## 6. Exact incomplete action (declared, NOT executed) + +Post-restart action, to be performed once, from the repository root, after +recovery context is established per section 9: + +``` +node --test scripts/test-rocko-launch.mjs +``` + +Why this one: it is the isolated fixture suite for the very seat being +restarted; it has not been run by me in this session (BUILD-LOG records another +actor's earlier pass, which I did not reproduce); and on inspection it has no +real runtime effects. Inspected behavior: it creates a temp directory under the +OS temp root, copies `agents/rocko` there, writes a fixture `scripts/common.sh`, +prepends a fake `claude` to PATH, drives the launcher through `script(1)` for a +PTY, exercises check, first launch, resume, failed-exit pointer retention, +fresh, corrupt pointer refusal, unknown-argument refusal, missing-user refusal, +non-TTY refusal, lock refusal, and user override, then removes the temp +directory. It never reads `~/.config/mosaic-dev`, the data root, or +`/mnt/storage/src/mosaic-stack/.pi/state`, and never calls a model. + +Expected output on success: a Node test reporter summary containing +`# tests 1`, `# pass 1`, `# fail 0`, process exit 0, and no change to +`git status` in the repository. Expected duration: a few seconds. + +Failure handling: +- Capture the complete stdout and stderr verbatim into the recovery artifact. +- Do not rerun, do not edit source or the test, do not mock or narrow. +- Classify from the output: environment (missing `script`, `flock`, or Node + feature; PTY unavailable in the restarted terminal) versus assertion failure + (a real launcher contract mismatch). Report the classification and the exact + failing assertion line to Darkwing. An environment failure is not evidence + against the launcher; an assertion failure is a finding for the trial's + independent review, not something I fix under this assignment. + +## 7. Outstanding or uncertain external operations + +- The CHECKPOINT READY announcement to Darkwing (socket `default`, target + `=darkwing`, class actionable). Its transport result is appended in section 10 + after the send. Exit 0 means dispatched only. No resend unless Darkwing or + Jason asks. +- No Gitea calls, no pushes, no worker dispatches, no subagents, no scheduled + wakeups, and no reply obligations to anyone other than Darkwing are outstanding + from this session. +- Unknown to me: whether Dewey's boundary declaration and Filbert's review + request have been delivered or answered. Not my action. + +## 8. Pause state + +Paused at CHECKPOINT READY, waiting for Jason's planned restart. The process +stays idle. Nothing is in flight. No timer, cron, loop, or self-restart is +registered. Aggregate usage is unavailable; no numeric budget was supplied. + +## 9. Restart recovery instructions (for the resumed Rocko) + +1. Confirm you were launched by `agents/rocko/launch.sh` and note whether the + launcher printed "resuming conversation" with the UUID in section 2 or + "starting conversation" with a new UUID (that happens with `--fresh` or a + lost history). Record which. If the UUID differs or this checkpoint is not + in your context, that is a normal-resume failure: record it first in the + recovery artifact before doing anything else, then proceed from files only. +2. Read, in order: root `AGENTS.md`, `docs/plans/CURRENT.md`, this checkpoint, + the inventory, and the task file. Do not ask Jason for a new briefing. +3. Verify: `git rev-parse HEAD` (expect `3b7fd19d…`; if different, record the + new head and do not assume the candidate is unchanged), `git status --short` + (expect the same untracked set plus the two artifacts; note any new + declared artifacts from Darkwing, Dewey, or Filbert), and that + `docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md` does not + yet exist. +4. Check for any message from Darkwing or Jason in your pane that revokes or + changes the assignment. Absence of a message is not authority to widen scope. + If authority cannot be established from these records, stop and report. +5. Write the recovery artifact header first: recovered checkpoint identity (this + file's path and its sha256 as found on disk), resume mode observed in step 1, + and the verification results of step 3. +6. Execute section 6 exactly once. Append the verbatim output and the + classification to the recovery artifact. Do not run anything else. +7. State remaining work honestly: independent review, publication, and owner + acceptance are not yours to claim. Then announce completion to Darkwing: + `tools/tmux/agent-send.sh -L default -s =darkwing -C actionable -m ""` + with the recovery artifact path. Transport exit 0 is not acknowledgement. +8. Session registration in `docs/SESSIONS.md` remains subject to Darkwing's + decision on inventory issue 5. If he has ruled, follow the ruling; if not, + ask in the completion message rather than writing. + +## 10. Delivery record + +(appended after the announcement is sent) + +Sent once at 2026-09-08 18:53 UTC: `tools/tmux/agent-send.sh -L default -s =darkwing +-C actionable` with the CHECKPOINT READY text (both artifact paths and hashes, the +declared unexecuted action, the ownership and SESSIONS.md decisions). Result: +exit 0, `transport dispatched; application acceptance unknown`. Delivery is +uncertain by contract. No resend. Darkwing's acknowledgement was not observed at +the time of writing. Appending this record changes this file's hash; the hash +quoted in the message (`c4374500…`) identifies the pre-send version, and the +post-send hash is whatever `sha256sum` reports on disk at recovery time. + +Correction (clock read `2026-09-08T18:46:09Z` immediately after the append above): +the send occurred at about 18:45 UTC, not 18:53. The "Written 18:50 UTC" header and +the inventory's "Prepared 18:45 UTC" were estimates made before reading the clock; +actual writes happened between 18:40 and 18:46 UTC. Ordering is unchanged: +inventory, checkpoint, send, delivery record, this correction. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md b/docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md new file mode 100644 index 000000000..e3d3ae163 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md @@ -0,0 +1,226 @@ +# PUB-REC-ROCKO-1 — publication inventory (read-only) + +Actor: Rocko (Claude Code, Fable). Coordinator: Darkwing. Owner: Jason. Trial #1497, +marker PUB-REC-20260908. Plan: `docs/plans/2026-09-08_publication-recovery-trial.md`. +Task: `docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md`. +Prepared 2026-09-08 18:45 UTC (2026-09-08 @ 13:45 CDT). + +Repository state at inspection: canonical checkout `/mnt/storage/src/mosaic-stack`, +branch `refactor`, HEAD `3b7fd19d087226b1887bf017fc10ef16b044f6aa` (equals the plan +baseline and `origin/refactor`). Index empty. No stash entries. + +This is publication review preparation, not independent approval. Nothing was +staged, edited, committed, pushed, launched, installed, or executed against real +runtime state. Private panes, transcripts, auth, and `.pi/state` contents were not +read; only directory names and my own launcher session pointer (a UUID) were listed. + +## Method + +- `git status --porcelain --untracked-files=all` and `git diff` for tracked changes. +- Full source read of every candidate file listed below. +- Syntax-only checks executed: `bash -n` on the five shell files and `node --check` + on the two test files. All passed. These write nothing. +- Pattern scan for credential shapes (counts only, no content printed): zero hits. +- Pattern scan for personal identifiers (home paths, email addresses, host names): + only the canonical checkout path appears, in seven files, and that path is + already published in tracked `AGENTS.md`. +- Not executed: `node --test`, any launcher (`--check` included), `scripts/skill.sh`, + Docker, Pi, or Claude model calls. + +## Inspected candidate paths + +Family A — shared launcher and entry point (tracked, modified): + +| Path | State | sha256 (working tree) | +| --- | --- | --- | +| `scripts/agent.sh` | tracked, +1 usage line | `ddfb43d2…d920145` | +| `scripts/agent-host-dev.sh` | tracked, +23/-3 | `07f40487…a336d5d1a` | +| `scripts/test-darkwing-launch.mjs` | tracked, +16/-11 | `030b4d09…7e39d9ab6` | +| `agents/darkwing/SOUL.md` | tracked, +11/-2 | `87e6ee74…9011c370783` | + +Family B — new agent seats and team directory (untracked): + +| Path | sha256 | +| --- | --- | +| `agents/README.md` | `f1ae5988b04cf1680d2eeefad23c119520ec365491c563237f7488242e10b925` | +| `agents/rocko/launch.sh` (mode 755) | `b65566e19fa0fb83409552954a645b8cdc4d2bdb1002df55d27088add6acf65d` | +| `agents/rocko/README.md` | `1576c37d8810231c65d3306c75667a632e5662062ee43f67cb43b01302e10ee5` | +| `agents/rocko/CONTEXT.md` | `e31c80e007d8b890c698048feb64c0f1fba9cf77f4836e007b9a9b7538b2d851` | +| `agents/rocko/SOUL.md` | `d5917650ad98ce6a9e51091d0758121db26877304491c8c00086675f28ffd26b` | +| `agents/filbert/launch.sh` (mode 755) | `6e5d04034e73d9dcacc6797a19b4e64c2d2e4575a3ebbab6ab3aadf86592681f` | +| `agents/filbert/README.md` | `d3bd85dd59142ca82d8b8a355921389dd1dd9da880c47b7a4760e85e557687e4` | +| `agents/filbert/CONTEXT.md` | `485737e40449a2fbc5215fa11357cd2706c3d90265d472a07aa9644d56230fec` | +| `agents/filbert/SOUL.md` | `8ab8142d0f118e9635d40e2a44d0497f6a6acf62deb50d5435715644b9c71e01` | +| `agents/filbert/validate-sessions.mjs` | `8a7dc4a3f368b0a695ac7d0f7da3f360dd12c7a231f5e7dbe6b576b80b409979` | +| `agents/dewey/launch.sh` (mode 755) | `b6a21d5ae195664612c1dd3f96670934f92025ea100c0e669fa4f7cacd5974f7` | +| `agents/dewey/README.md` | `908a5e6e9eea2be2b190ba87a7dade4fa6228bea21aec0b09746a5176fdeda06` | +| `agents/dewey/CONTEXT.md` | `61d640098d8ad79c9c069ea761c8a50c5f644cb914a253fe037132c61a4c421c` | +| `agents/dewey/SOUL.md` | `b156fd7c76dfe7c8eb421aeed9d3fd1b681456ca96961d5a010cbe8e1ecf0e33` | +| `agents/dewey/validate-sessions.mjs` | `d0136b89509edff48aa0e2ee398acb7cc7ccd47255225823e47892c57384e60b` | + +Full hashes for family A: `scripts/agent.sh` +`ddfb43d2599a722452376eee52cbc66da250722f0de55be1570eb55c4d920145`; +`scripts/agent-host-dev.sh` +`07f4048729c36c57b6546d3776c81ff85dd89af93666a15b7c41805a336d5d1a`; +`scripts/test-darkwing-launch.mjs` +`030b4d094c354085dd1d98a644a852d949a4f4ed1cba0328ef8a1907e39d9ab6`; +`agents/darkwing/SOUL.md` +`87e6ee7459f705cfb0c354cccade74280feb05408187c52cbf92b9011c370783`. + +Family C — tests and skill (untracked): + +| Path | sha256 | +| --- | --- | +| `scripts/test-rocko-launch.mjs` | `13c32cad02587aad85da2f1a2bca8e61d77d9285ffc9e21dc91b7bafa7d0e05d` | +| `skills/ms-executive-update/SKILL.md` | `368a857632dfd118a11a752ae0a775ca13d6837cf05d8f870946c9e7364d127f` | + +Family D — trial records (untracked, this trial's own paperwork): + +| Path | sha256 | +| --- | --- | +| `docs/plans/2026-09-08_publication-recovery-trial.md` | `ab90830848a6fdeb89c0baffdddb7d28343ae4fe4436873a3d62f7611282d328` | +| `docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md` | `af673b60e37831f6ffd21fffe76a831696106798827925cd904b6a4472090603` | + +Also present but identified only, not content-reviewed for publication (owned by +other actors, append-only or in flight): `BUILD-LOG.md` (+36), `docs/SESSIONS.md` +(+11), `docs/plans/CURRENT.md` (+15/-3), `docs/plans/reviews/2026-09-08_publication-trial-start-receipt.md`, +`docs/plans/reviews/2026-09-08_publication-trial-dewey-request.md`, and the untracked +`docs/plans/reviews/2026-09-07_*` wave records listed at session start. + +## Excluded from inspection + +- `agents/dewey/work/**` (about 5.8 MB: WUI drafts, bundled fonts, logos, PNG + evidence, check scripts) and `agents/dewey/scratch/` (empty). Dewey declares that + boundary under PUB-REC-DEWEY-1. Not read. +- `.pi/state/**` (per-agent launch snapshots, session JSONL, locks). Ignored by the + tracked `.pi/.gitignore` (`state/`, `evidence/`, `extensions/`), confirmed with + `git check-ignore`. Launch snapshots embed the live `USER.md`, so this exclusion is + publication-critical and currently holds. +- Any auth file, transcript, or other agent's pane. +- `v1/` legacy archive. + +## Dependencies + +Pi seats (Darkwing, Dewey, Filbert): `agents//launch.sh` execs +`scripts/agent.sh --host-dev `, which execs `scripts/agent-host-dev.sh`. That +helper depends on `scripts/common.sh` (`load_config`, adapter must be `pi`), +`node_modules/.bin/pi` matching the `package.json` pin 0.84.4, `flock`, +`scripts/sync-dev-extensions.sh`, `extensions/goal/index.ts`, +`.pi/extensions/goal/index.ts`, ten skill directories (`ms-tools`, `ms-file-read`, +`ms-file-write`, `ms-conductor`, `ms-communications`, `ms-agent-watch`, `ms-sdlc`, +`ms-proactive-agent`, `ms-goal`, `ms-unslop`), `agents//{SOUL,CONTEXT}.md`, +`agents//validate-sessions.mjs`, `contracts/{CONSTITUTION,STANDARDS}.md`, +root `AGENTS.md`, and `/user/USER.md`. Filbert's wrapper appends +`--provider openai-codex --model gpt-6-astra --thinking low` after user arguments. +The three `validate-sessions.mjs` copies differ only in the agent name inside the +error string. + +Claude seat (Rocko): `agents/rocko/launch.sh` depends on `scripts/common.sh` +(`load_config` for the data root only), `claude` on PATH, `node`, `flock`, the same +contracts, `AGENTS.md`, `agents/rocko/CONTEXT.md`, and `/user/USER.md`. +It records a UUID in `.pi/state/rocko/session-id` and resumes it. Observed evidence +that it works live: the session pointer equals this process's own session UUID +(`19240012-81d2-4005-81f6-a48c4b8557fc`), and `.pi/state/rocko/launches/` holds one +receipt. + +Tests: `scripts/test-rocko-launch.mjs` builds a temp root, copies `agents/rocko`, +writes a fixture `scripts/common.sh`, prepends a fake `claude` to PATH, and uses +`script(1)` for a PTY. `scripts/test-darkwing-launch.mjs` builds a temp root per +agent, copies `agents/` recursively, `scripts/agent-host-dev.sh`, +`scripts/agent.sh`, and the whole `skills/` tree, then stubs `common.sh`, +`sync-dev-extensions.sh`, `package.json`, the goal extension, and a fake `pi`. A +fourth test stubs Docker for the container entry point. Neither test touches the +real data root, config, `.pi/state`, or a model. Both need `script`, `flock`, and +Node with `import.meta.dirname` (host has Node 26.8.1). + +Skill: `skills/ms-executive-update/SKILL.md` is standalone Markdown. It is not in +the host-dev helper's ten-skill list, so no launcher loads it. A separate, +tracked, deliberately historical copy lives at +`extensions/goal/test/fixtures/skills-local/ms-executive-update/SKILL.md` and is +consumed by the goal extension tests; the two files differ from byte 1148. + +## Ownership uncertainty + +- `docs/SESSIONS.md` attributes all three 2026-09-08 entries for this work + (skill sync 12:18 UTC, Dewey agent 12:48 UTC, Rocko/Filbert/team lead 12:53 UTC) + to actor "Codex". No seat named Codex exists under `agents/`, and the tracked + bases of family A were last committed in `7c580a56` under the git user Jason + Woltje. The 2026-09-07 wave inventory explicitly lists every family A, B, and C + path as "newer relocation work: do not adopt, stage, revert or clean without + ownership reconciliation." Darkwing should confirm who owns these files before + any staging. I hold no ownership of them; my own seat files under + `agents/rocko/` were also authored by that actor, not by me. +- `agents/darkwing/SOUL.md` is Darkwing's persona, modified by that same actor. + Darkwing has not, in the records I can see, accepted the change to his own SOUL. +- The SESSIONS claim that the skill is "byte-identical to operator local skill" + cannot be re-verified: no `ms-executive-update` directory exists under + `~/.config/mosaic/skills` or `~/.claude/skills` now. +- `agents/dewey/SOUL.md`, `CONTEXT.md`, and `README.md` describe Dewey's role and + were authored by Codex, not Dewey. Dewey's boundary declaration should say + whether he accepts them as his. + +## Publication-sensitive content categories (no content exposed) + +1. Absolute canonical checkout path in seven candidate files. Already public in + tracked `AGENTS.md`; no new exposure. The tracked baseline already has 31 files + containing a home-directory path, so this is an existing publication posture, + not a new one. +2. Model, provider, and vendor identifiers (Fable, OpenAI Codex GPT-6 Astra, + Claude Code, Pi) in personas, READMEs, and tests. Not secrets. +3. Personal user profile content: none in the candidates. It enters only the + ignored `.pi/state` launch snapshots at runtime. +4. Credentials or tokens: none found by pattern scan. Launchers explicitly use + native authentication and copy nothing. +5. Team member names and the owner's first name appear in personas and plans. + Consistent with already-published `AGENTS.md`. +6. Third-party or licensed material: none in these families. Fonts and logos are + confined to the excluded Dewey subtree. + +## Suggested applicable tests (isolated, no runtime or model effects) + +| Check | Command from repo root | Effects | Status by me | +| --- | --- | --- | --- | +| Rocko launcher contract | `node --test scripts/test-rocko-launch.mjs` | temp dir only; fake `claude` | NOT RUN (declared post-restart action) | +| Pi seat launchers + container entry | `node --test scripts/test-darkwing-launch.mjs` | temp dirs only; fake `pi`, fake `docker` | NOT RUN | +| Shell syntax | `bash -n` on `scripts/agent.sh`, `scripts/agent-host-dev.sh`, three `agents/*/launch.sh` | none | RUN, pass | +| Test-file syntax | `node --check` on both `.mjs` tests | none | RUN, pass | + +Not recommended for review execution: real `agents/*/launch.sh --check` (runs real +`claude --version` or pinned `pi --version` and loads real config; harmless but +executes a launcher against real state, which this task forbids), and +`scripts/skill.sh install` (writes to the data root). Suites `test-config.sh`, +`test-auth.sh`, `test-conductor.sh`, `test-task.sh`, and `test-release.sh` do not +exercise these files. The goal extension tests read the fixture skill copy, not the +candidate, so they are unaffected by the new skill file. + +## Issues found + +1. Documentation gap: `docs/TOOLS.md` line 64 shows the `--host-dev` usage without + the new `--provider`, `--model`, and `--thinking` flags that + `scripts/agent-host-dev.sh` and `agents/README.md` now document. Low severity; + fix before or alongside publication. +2. Dangling references in `skills/ms-executive-update/SKILL.md`: it cites + `$MOSAIC_BRAIN_HOME`, `fleet/lanes//TASKS.md`, `fleet/agents//`, + `fleet/board/`, `docs/guides/FLEET-COMMS.md`, and `docs/STRUCTURE-CANON.md`. + None of those paths exist in this repository. The skill is a copy of an + operator-side convention; publishing it as-is documents a layout this + repository does not have. Provenance pointer is also unverifiable (see + ownership). Suggest a provenance note or a pending label if published. +3. Test coupling to an active author's subtree: `scripts/test-darkwing-launch.mjs` + copies `agents/dewey` recursively, which now includes about 5.8 MB of WUI + drafts, binary fonts, and evidence PNGs plus `scratch/`. Assertions do not use + that content, but the copy runs while Dewey may be writing, and the fixture + grows with his work. Low severity; an exclusion filter would decouple it. +4. `agents/rocko/launch.sh` passes the whole context snapshot as one argv value + (`--append-system-prompt "$(cat …)"`), unlike the Pi helper which passes a path. + Fine at the current size; it will hit argv limits only if the inputs grow + greatly. Note only. +5. Session registration conflict: root `AGENTS.md` requires every session to + append to `docs/SESSIONS.md`, but PUB-REC-ROCKO-1 instructs me to write only + the three declared artifacts. I followed the task instruction and did not + append. Darkwing should decide whether the registration is written by him, + by me after restart, or as part of the recovery artifact. +6. No correctness defect was found in the launcher or test logic on reading. + Argument validation, fail-closed context checks, the launch lock, UUID + validation, session-file validation, and the thinking-level allowlist all + read as intended. This is a source reading, not an execution result. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md b/docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md new file mode 100644 index 000000000..5fdd47040 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md @@ -0,0 +1,129 @@ +# PUB-REC-ROCKO-1 — post-restart recovery record (Rocko) + +Trial marker: PUB-REC-20260908 (#1497). Plan: `docs/plans/2026-09-08_publication-recovery-trial.md`. +Task: `docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md`. +This file is the declared recovery artifact reserved by checkpoint section 9. It is +append-only; corrections are new paragraphs, never edits. Written by Rocko (Claude Code, +Fable) on the host in the canonical checkout. + +## 1. Recovered checkpoint identity + +- Checkpoint: `docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md` +- sha256 as found on disk after restart: `30c4e1906b22220a29868c1ef4253676d2e2324d46a1f281543e50347af7fdd8` + (matches the post-correction hash recorded by Darkwing's readiness ruling; the + pre-send hash `c4374500…` recorded inside the checkpoint's own delivery paragraph is + superseded by the appended timestamp correction, as that paragraph explains). +- Inventory: `docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md`, + sha256 `51536b1d4a8fee30caa711527295ed7b860d46ae9b81e5530af6631a53030b5f` (unchanged). +- Coordinator ruling read and applied: `docs/plans/reviews/2026-09-08_publication-trial-checkpoint-readiness.md`, + sha256 `6ce6109a77fa46f1dff7e4a866b9fc1518dbe959f20bdd5b83c7669e8aadcc14`. +- Operator restart receipt read: `docs/plans/reviews/2026-09-08_publication-trial-operator-restart-receipt.md`, + sha256 `daca478a3ec4a564c3b3b87c7fc57d14826166762e745ad6b9375da9e659d6f6`. + +## 2. Resume mode observed (section 9 step 1) + +- The process that writes this file has the same hosted session link recorded in + checkpoint section 2 and holds the prior conversation (the pre-restart task, + ruling, checkpoint writes, and CHECKPOINT READY send are all in context). That is + consistent with a normal `--resume` of the recorded session ID via + `agents/rocko/launch.sh` with no `--fresh`. +- The launcher's own stdout announcement of the resume mode is not observable from + inside the resumed process, and per the coordinator ruling no `.pi/state` or other + private-state reads were performed after restart. Resume mode is therefore inferred + from retained context, not read from the pointer file. +- Normal-resume failure: none observed. No assisted recovery (re-briefing) was needed + or received. The only post-restart input was the pre-authorized neutral `continue`. +- The hosted session was context-compacted once after restart (a harness summary + replaced the earlier transcript). Task facts used below were re-read from the + declared on-disk records, not taken from the summary alone. + +## 3. Pre-execution verification (section 9 steps 3–4), clock 2026-09-08T18:52:18Z + +| Check | Result | +| --- | --- | +| HEAD | `3b7fd19d087226b1887bf017fc10ef16b044f6aa` on `refactor` (matches checkpoint §3) | +| Index | empty (no staged changes) | +| `git status --short` | 27 lines, saved for post-run comparison | +| Recovery artifact | absent before this write | +| `scripts/test-rocko-launch.mjs` | `13c32cad02587aad85da2f1a2bca8e61d77d9285ffc9e21dc91b7bafa7d0e05d` (unchanged) | +| `agents/rocko/launch.sh` | `b65566e19fa0fb83409552954a645b8cdc4d2bdb1002df55d27088add6acf65d` (unchanged) | +| Tooling | node v26.8.1, `/usr/bin/script`, `/usr/bin/flock` present | +| `docs/plans/CURRENT.md` next action | still the #1497 trial; PUB-REC-ROCKO-1 named; C1 HELD | +| Revoking messages | none received; only `continue` | + +New records from other actors since the checkpoint (not inspected beyond filenames and +the operator receipt): `…-operator-restart-receipt.md`, `…-dewey-boundary.md`, +`…-wui-coordination.md`, `…-wui-review-request.md`, `…-wui-snapshot.json`, +`…-wui.sha256`. `docs/SESSIONS.md` and `BUILD-LOG.md` now show as modified in the +working tree; Darkwing owns those registrations, and Rocko did not touch them. + +## 4. Declared bounded action (checkpoint section 6), executed exactly once + +Command, run from the repository root: + +``` +node --test scripts/test-rocko-launch.mjs +``` + +Expected: `# tests 1`, `# pass 1`, `# fail 0`, exit 0, `git status --short` unchanged. + +### 4.1 Execution record + +- Started: 2026-09-08T18:52:51Z; ended: 2026-09-08T18:52:51Z; wall time under one second. +- Process exit code of `node --test`: 0. +- Run count: one. No rerun. + +Verbatim stdout: + +``` +✔ Rocko uses Fable, isolated resume IDs, current context and launch refusals (666.457476ms) +ℹ tests 1 +ℹ suites 0 +ℹ pass 1 +ℹ fail 0 +ℹ cancelled 0 +ℹ skipped 0 +ℹ todo 0 +ℹ duration_ms 706.5949 +``` + +Verbatim stderr: empty (zero bytes). + +### 4.2 Classification + +- Result: PASS. Assertions match the declared expectation (1 test, 1 pass, 0 fail, exit 0). + Node's reporter printed the summary lines with an `ℹ` prefix instead of the `#` prefix + quoted in section 6 of the checkpoint; that is reporter formatting, not a deviation in counts. +- Working tree: `git status --short` after the run differs from the pre-run snapshot by exactly + two untracked lines, neither produced by the test: + 1. `docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md` (this file, written in + step 5 after the pre-run snapshot was taken). + 2. `docs/plans/reviews/2026-09-08_publication-trial-wui-condition-disposition.md` (a new + file from another actor that appeared between the two snapshots; not inspected by Rocko). + No tracked file changed and nothing was staged. The test used an isolated temp root and + cleaned it up, as its source declares. +- Environment vs assertion: no failure of either kind to classify. + +## 5. Remaining work and what this record does not claim + +- This record shows only that Rocko recovered PUB-REC-ROCKO-1 from the declared checkpoint + after a planned restart and executed the one reserved read-only action with the expected + result. It does not claim the whole #1497 trial accepted. +- Not done by Rocko and not Rocko's to do: independent review of the recovery, curated + publication derivative of the raw checkpoint/inventory, SESSIONS/BUILD-LOG registration + (Darkwing owns), the separate WUI review, and owner acceptance by Jason. +- Archify C1 remains HELD. No source, test, or launcher edits; no commits, pushes, launches, + or private-state reads were performed after restart. +- Inventory issues (docs/TOOLS.md line 64 flag gap, skill dangling references, test copying + the Dewey subtree, argv-size note) remain open for Darkwing's disposition. + +## 6. Delivery record + +- Announced once to Darkwing at 2026-09-08T18:53:24Z via + `tools/tmux/agent-send.sh -L default -s =darkwing -C actionable`, pointing at this file with + its section-5 hash `52778f1a681277335f5d86a1e1fe2b96e9df07303b81c1d833ed674cf44811f7` + (the hash of this file before this delivery paragraph was appended). +- Transport result: exit 0, "transport dispatched; application acceptance unknown". Exit 0 is + not an acknowledgement. No resend will be attempted; if Darkwing reports non-receipt, the + pointer above is the reference. +- Rocko is idle after this write. No further action is queued. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md b/docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md new file mode 100644 index 000000000..3c77ae8be --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-rocko-task.md @@ -0,0 +1,20 @@ +# PUB-REC-ROCKO-1 — inventory and planned checkpoint + +Owner Jason authorized the publication/recovery trial; coordinator Darkwing. Read `docs/plans/2026-09-08_publication-recovery-trial.md`. Canonical checkout /mnt/storage/src/mosaic-stack, refactor. Separate bounded assignment; Archify C1 stays HELD. If another active assignment conflicts, report conflict before proceeding. + +## Task before operator restart + +Read-only inspect pending general launcher/team-definition/test/skill changes, excluding Dewey's WUI subtree and private session/auth/runtime files. Use git diff for tracked changes; inspect source files before drawing conclusions. Candidate families: scripts/agent.sh, scripts/agent-host-dev.sh, scripts/test-darkwing-launch.mjs, scripts/test-rocko-launch.mjs; agents team README, SOUL/CONTEXT/launch/README/validate-sessions code; skills/ms-executive-update/SKILL.md. Do not change source, stage, commit, push, deploy, start a model process, or execute a launcher against real state. Never print credentials or inspect private panes/transcripts. This is publication review preparation, not independent final approval. + +Write ONLY these shared task artifacts: +- `docs/plans/reviews/2026-09-08_publication-trial-rocko-inventory.md`: inspected paths, dependencies, ownership uncertainty, publication-sensitive content categories without exposing content, suggested applicable tests and any issues. +- `docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md`: assignment/authority, settled decisions, evidence pointers, exact incomplete action, outstanding/uncertain external operations, pause state, and restart recovery instructions. +- `docs/plans/reviews/2026-09-08_publication-trial-rocko-recovery.md`: reserve for AFTER Jason restarts you; do not create completion evidence now. + +Set the declared post-restart next action to a concrete read-only verification you have NOT yet performed, based on inspected test code. If possible choose an isolated fixture suite with no real runtime/model effects; inspect its behavior before recommending it. State expected outputs and failure handling. Leave this action unexecuted before restart. + +Then announce CHECKPOINT READY via agent-send.sh to Darkwing on default socket =darkwing, with checkpoint path, and wait. No automatic self-restart. Jason alone performs the planned restart after coordinator checks. A transport exit 0 is not ACK. Record uncertain delivery without blindly resending. + +## After restart + +Recover this task via existing session/declared records, not a new briefing from Jason. Record recovered checkpoint identity and execute the declared next bounded action exactly once only after owner restart/resume context is clear. Capture the result and remaining work in the recovery artifact; do not claim the whole trial accepted. If recovery cannot establish authority, stop and report rather than guess. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-sonnet-r2-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-sonnet-r2-review-request.md new file mode 100644 index 000000000..73cb12a83 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-sonnet-r2-review-request.md @@ -0,0 +1,7 @@ +# PUB-REC-FILBERT-SONNET-2 + +SONNET-F1/F2 corrected with three Fable-to-Sonnet wording replacements in agents/rocko/CONTEXT.md, agents/README.md and agents/darkwing/SOUL.md. No authority/persona behavior changed or live session restarted. Coordinator missed these references in round 1; original rejected verdict remains unchanged. Historical Fable recovery evidence remains unchanged. + +Coordinator verification: node --test scripts/test-rocko-launch.mjs scripts/test-darkwing-launch.mjs, five passed, zero failed; git diff --check clean. Fake model fixtures only, no actual model calls. + +Filbert: review frozen seven-file snapshot /tmp/pub-rec-sonnet-r2-htbgwodg, SHA256SUMS manifest c5ad630615f67c4480c23f3478123c7c1c5e7c6095bcfa45039650c8213c38bf. Compare against round-1 snapshot /tmp/pub-rec-sonnet-review-u__t4rsb and original approved source snapshot /tmp/pub-rec-launcher-review-a35yzs12 for the two team files. Exactly six changed candidate files from original source approval; Rocko SOUL is unchanged fixture context. Confirm model wording across those candidates, minimal deltas, hash integrity and test in disposable copy only. No private state, live launch, configuration change, source edits or deployment. Write only docs/plans/reviews/2026-09-08_publication-trial-sonnet-r2-verdict.md. Return verdict to Darkwing; publication review only, not trial acceptance. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-sonnet-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-sonnet-review-request.md new file mode 100644 index 000000000..ba215939c --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-sonnet-review-request.md @@ -0,0 +1,5 @@ +# PUB-REC-FILBERT-SONNET-1 + +Jason changed Rocko's launcher to Sonnet and authorized matching documentation repairs. Darkwing changed only README model wording and the isolated test's title/expected model. Coordinator test: one pass, zero failures; no real model call. Historical recovery records and frozen prior snapshots remain unchanged. Earlier publication approval does not cover these revised bytes. + +Filbert: independently review the five-file snapshot at /tmp/pub-rec-sonnet-review-u__t4rsb, manifest SHA-256 52a1c408aad20b8c71377db2e456a11450b1d72264adaa679a26f9374c48e263. Compare the three changed files against /tmp/pub-rec-launcher-review-a35yzs12. Read-only review; run the isolated test only in a disposable copy after inspecting isolation. No actual launch, private-state reads, configuration edits or deployment. Verify model/docs/test consistency and no unrelated changes. Write only docs/plans/reviews/2026-09-08_publication-trial-sonnet-verdict.md and report to Darkwing. Publication review only; no trial acceptance implied. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-start-receipt.md b/docs/plans/reviews/2026-09-08_publication-trial-start-receipt.md new file mode 100644 index 000000000..6e600a8a8 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-start-receipt.md @@ -0,0 +1,9 @@ +# Publication/recovery trial kickoff receipt + +Timestamp: 2026-09-08 UTC. Baseline 3b7fd19d, refactor. Owner confirmed plan after four interview rounds. Plan: docs/plans/2026-09-08_publication-recovery-trial.md. + +Issue created once: HTTP 201, client exit 0, #1497 https://git.mosaicstack.dev/mosaicstack/stack/issues/1497. Marker PUB-REC-20260908. No duplicate POST. + +Live session-existence metadata only: Rocko on mosaic-fleet; Filbert and Dewey on default. No private panes inspected. PUB-REC-ROCKO-1 request sent once to =rocko on mosaic-fleet, sender dragon-lin:darkwing. Exit 0, `transport dispatched; application acceptance unknown`. Assignment acknowledgement and checkpoint readiness NOT yet observed. No automatic timer or restart. No blind resend. + +Outstanding expected artifact: docs/plans/reviews/2026-09-08_publication-trial-rocko-checkpoint.md, with inventory and a still-unexecuted post-restart action. Coordinator must check readiness, then invite Jason to restart. No planned-restart success or publication claimed. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-wui-coordination.md b/docs/plans/reviews/2026-09-08_publication-trial-wui-coordination.md new file mode 100644 index 000000000..3f54bd714 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-wui-coordination.md @@ -0,0 +1,7 @@ +# WUI boundary admission and independent review coordination + +PUB-REC-DEWEY-1 author reply received. Coordinator read the complete declaration and verified the 53-file manifest and every source/copy hash. Snapshot and admission receipt are recorded in 2026-09-08_publication-trial-wui-snapshot.json. No source modified; exclusions and provisional-default/feedback distinction preserved. + +One terminal-log acknowledgement sent to Dewey on default socket =dewey, sender dragon-lin:darkwing. Exit 0, transport dispatched/application acceptance unknown. No reply requested. + +One actionable independent-review request PUB-REC-FILBERT-WUI-1 sent to Filbert on default socket =filbert, same sender. Exit 0, transport dispatched/application acceptance unknown; reviewer assignment acknowledgement/approval not yet observed. Request 2026-09-08_publication-trial-wui-review-request.md pins the separate snapshot, not mutable author files. Expected verdict: 2026-09-08_publication-trial-wui-verdict.md. Coordinator owns reconciliation; no blind resend or timer. Draft publication review does not grant product/branding acceptance. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-wui-review-request.md b/docs/plans/reviews/2026-09-08_publication-trial-wui-review-request.md new file mode 100644 index 000000000..d45afc4fe --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-wui-review-request.md @@ -0,0 +1,19 @@ +# PUB-REC-FILBERT-WUI-1 — independent draft publication review + +Reviewer requested: Filbert. Author: Dewey. Coordinator: Darkwing. Owner-approved trial #1497, docs/plans/2026-09-08_publication-recovery-trial.md. This is a bounded review, not authorship or implementation. If a conflicting task or prior authorship prevents independence, report it before acting. + +## Exact candidate + +Read the author declaration `docs/plans/reviews/2026-09-08_publication-trial-dewey-boundary.md` and admission receipt `docs/plans/reviews/2026-09-08_publication-trial-wui-snapshot.json`. + +Review ONLY the separate read-only snapshot `/tmp/pub-rec-wui-review-cgfkxmll`, not the live WUI tree. Verify its SHA256SUMS before admission. Manifest SHA-256 `a585c1eedb17044962c0419861a1032b23409b769c5d304aceff6ef6089a0e11`, 53 paths. Durable manifest also at docs/plans/reviews/2026-09-08_publication-trial-wui.sha256. Copy to a disposable directory before any checks that regenerate evidence; never mutate the snapshot or author's files. + +## Required review + +- Confirm exact allowed source/assets/evidence and actual HTML/CSS/JS/font dependencies form a usable standalone draft. Do not add excluded files to satisfy historical documentation references. +- Inspect publication safety of curated notes, structured evidence and allowlisted screenshots, including sensitive material and unintended private data. Never inspect auth/private runtime/session files. Report sensitive findings by category/location without quoting secrets. +- Review third-party font provenance and accompanying OFL notices, including reserved-name boundaries. No copying licenses from archived v1 or claiming trademark clearance. +- Preserve draft status: original board defaults remain provisional. Jason liked themes/Manrope; those source/default changes are not implemented. Logo completion no longer blocks the separate dashboard-design direction, but dashboards have not started. Publication must not imply branding/product acceptance. +- Check that claims match evidence; distinguish author's browser checks from fresh independent reruns. Inspect helpers before executing; no network fetch, model invocation, real browser profile/state or runtime deployment. Reproduce isolated offline checks as useful, or explicitly state bounded NOT RUN limits. This is review for publishing a draft, not full product certification. + +Write ONLY `docs/plans/reviews/2026-09-08_publication-trial-wui-verdict.md`, with exact manifest, findings, checks and approval/request-changes for draft publication. Report any needed label/addendum rather than editing author source. Preserve hash identity and recheck it after review. No git operations, live-seat mutation, source changes or operator restart. Reply with verdict pointer via agent-send.sh to default socket =darkwing; exit 0 is transport only. diff --git a/docs/plans/reviews/2026-09-08_publication-trial-wui-snapshot.json b/docs/plans/reviews/2026-09-08_publication-trial-wui-snapshot.json new file mode 100644 index 000000000..d43220041 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_publication-trial-wui-snapshot.json @@ -0,0 +1,12 @@ +{ + "task": "PUB-REC-DEWEY-1", + "manifestSha256": "a585c1eedb17044962c0419861a1032b23409b769c5d304aceff6ef6089a0e11", + "fileCount": 53, + "sourceHashesVerified": true, + "snapshotHashesVerified": true, + "snapshot": "/tmp/pub-rec-wui-review-cgfkxmll", + "boundarySha256": "d6ac28eaf5a585415421610bde1f4fa850953ae7a5e48944f4675b090ebf5dad", + "status": "author-boundary admitted; independent publication review pending", + "excluded": "all paths not explicitly listed in author manifest", + "designStatus": "original provisional defaults remain; owner feedback is not source implementation or branding acceptance" +} diff --git a/docs/plans/reviews/2026-09-08_skill-launcher-coordination.md b/docs/plans/reviews/2026-09-08_skill-launcher-coordination.md new file mode 100644 index 000000000..c5e251ef9 --- /dev/null +++ b/docs/plans/reviews/2026-09-08_skill-launcher-coordination.md @@ -0,0 +1,5 @@ +# #1498 independent review wait + +SKILL-LAUNCHER-FILBERT-1 dispatched to Filbert/default in canonical checkout using agent-send.sh. Transport exit 0, application acceptance unknown. Required return: docs/plans/reviews/2026-09-08_skill-launcher-verdict.md plus reviewer message. Darkwing owns follow-up. Concrete wake path is the existing agent-message return; no timer or polling is armed. + +After dispatch, all nine frozen/live file hashes and the explicit old-skill deletion matched the declared candidate. Index untouched. Nearest gate: independent exact-candidate approval before selective commit/push. Other untracked files remain outside this assignment. No further source edits while pinned; on return verify verdict and identities before integration. No independent authorized implementation task remains before this review gate. diff --git a/docs/plans/reviews/2026-09-08_skill-launcher-review-request.md b/docs/plans/reviews/2026-09-08_skill-launcher-review-request.md new file mode 100644 index 000000000..22dee00dc --- /dev/null +++ b/docs/plans/reviews/2026-09-08_skill-launcher-review-request.md @@ -0,0 +1,11 @@ +# SKILL-LAUNCHER-FILBERT-1 (#1498) + +Owner goal: fix skill/launcher mismatch autonomously through reviewed commit and push; preserve other work; no live restarts or deployment. Darkwing author, Filbert independent reviewer. Declare any conflict before acting. + +Review frozen nine-file candidate /tmp/skill-launcher-review-5gh6v4v0 at SHA256SUMS c4dd2ad8ec87138fc756aa3e7a3366dcec32fa8c5d1ecfa3736f746457568904 plus explicit deletion skills/ms-agent-watch/SKILL.md. Baseline 12ff5da7df91f5343a59999af98b67518c15d6a4. Exact metadata in docs/plans/reviews/2026-09-08_skill-launcher-review.json is a separate publication receipt. No other untracked skills or private/WUI records are candidates. Preserve original source/provenance: replacement ms-watch existed before this assignment; Darkwing narrowed corrections to actual CLI compatibility/transport claims, not watch runtime behavior. The owner goal authorizes reconciling that replacement, not adopting unrelated new skills. + +Check launcher selects ms-watch and fails closed if it is absent; new offline regression exercises all three Pi agents, exact new skill argument and absent old argument. Missing skill creates neither .pi/state nor fake Pi capture. Communications/proactive skill references renamed only. Read replacement against tools/agent-watch/agent-watch.sh from baseline: explicit sender override, socket precedence, query side effects, user-bus refusal and legacy transport logs honestly qualified. No activation or reliability claims. Legacy watcher implementation and other historical stale statements are not silently rewritten. + +Reproduce in a writable disposable git-archive baseline plus exact overlay and declared deletion, never in the drifted live tree. Allowed: isolated node --test scripts/test-darkwing-launch.mjs scripts/test-rocko-launch.mjs, bash -n, static/hash checks. Inspect fixtures before execution. Coordinator observed failing baseline then five passing tests after correction. No actual launcher, model call, timer/systemd command, watch/list/status call, private pane/session/config reads, source edits or git writes. Snapshot remains unchanged. Do not append shared logs while frozen. + +Write ONLY docs/plans/reviews/2026-09-08_skill-launcher-verdict.md and reply through agent-send.sh to Darkwing/default. Return APPROVED or REQUEST CHANGES for exact source and tracking candidate with independent test evidence, privacy/scope and any contract concerns. Goal is reviewed commit/push, not new deployment or operator acceptance of unrelated work. diff --git a/docs/plans/reviews/2026-09-09_bma-foundation-pointer-handoff.md b/docs/plans/reviews/2026-09-09_bma-foundation-pointer-handoff.md new file mode 100644 index 000000000..8779558c7 --- /dev/null +++ b/docs/plans/reviews/2026-09-09_bma-foundation-pointer-handoff.md @@ -0,0 +1,15 @@ +# BMA foundation pointer handoff + +Request BMA-FOUNDATION-POINTERS-VELMA-7B859C4E1. Velma owns BMA/RoleCraft delivery; this response supplies existing foundation references only. No connector, login, breakglass or endpoint design is commissioned here. + +Authoritative versioned source: mosaicstack/stack, refactor commit 50d2a2eba9c93d3f801017ea6fd53a30308d9ab1. Canonical checkout /mnt/storage/src/mosaic-stack. The following files were checked against that commit and match live bytes: + +- AGENTS.md: root foundation, v1 archived; fail-closed configuration/policy; mission/task least-privilege intersection; immutable run evidence; credentials runtime-only, not repository/image content. Native host development agents are not sandboxed workers or managed fleet registrations. +- docs/TOOLS.md: existing integration-facing contract is the documented local CLI/file surface: task validation/execution, role resolution, immutable runs, release/auth commands. This pointer does not establish a hosted HTTP API or a BMA connector interface. Do not treat internal runtime data paths as a public API. +- docs/plans/2026-09-03_auth-provider-harness-registry.md: DRAFT FOR OWNER REVIEW, no implementation authorized. Separates harness, provider, account, endpoint and materialization. Its agent/provider credentials are not an implemented customer/public-login, Authentik or application-breakglass contract. Do not depend on proposed registration fields or materialization endpoints as shipped behavior. +- docs/plans/2026-09-06_workspace-schema-and-audit.md: proposed schema/audit detail, not an implementation contract. Owner-discussed boundaries include explicit project/workspace registration, least-privilege narrowing, scoped identity and no implicit cross-workspace access. Exact fields/mechanisms remain proposed. Future integration should avoid assumptions that a display name or login alone grants project/workspace access. +- docs/plans/reviews/2026-09-07_foundation-inspector-demo.md plus docs/plans/reviews/2026-09-07_a9-owner-acceptance.md: accepted offline synthetic permission/scope inspector only. Preview output is not a live grant, enforcement service or security certification. The latter supersedes the demo's historical pending-owner-acceptance status. + +No current commercial BMA HTTP endpoint, Authentik/public-login or independent app-breakglass contract was established by these sources. That is a bounded source finding, not a claim about all other teams' projects. Source-specific BMA work can continue independently; do not infer a foundation implementation dependency from these drafts. + +Current remit remains registry review alignment only. Live CURRENT.md has additional uncommitted text relative to the source pin; its leading next-action block was re-read and agrees with that remit, but this handoff does not claim the whole live file equals the pin. No unrelated edits adopted and no private state or credentials read. diff --git a/docs/plans/reviews/2026-09-10_m20-correction-review-request.md b/docs/plans/reviews/2026-09-10_m20-correction-review-request.md new file mode 100644 index 000000000..4e0923eba --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-correction-review-request.md @@ -0,0 +1,11 @@ +# M20-CORRECTION-FILBERT-1 (#1500) + +Jason approved the narrow D1-D10 corrective slice, not materialization. Author Darkwing; independent reviewer Filbert. Disclose original I1 review role and verify fresh evidence, not inherited approval. + +Review frozen 14-file package/progress snapshot /tmp/m20-correction-review-5ffad9u7, SHA256SUMS af97b5be51760f9075e6df859fee8cedf068fbb6b8464f208b48c129f1f99991. Manifest metadata: docs/plans/reviews/2026-09-10_m20-correction-snapshot.json. Baseline b3fa221060abb820cd809f15c3a54ba84518b7a1. Only package source/tests/fixtures/README change; the package manifest and index are unchanged contextual files. No root pin, dependency or other-agent work adopted. + +In writable disposable git-archive baseline plus overlay, rerun package tests and offline Pi/Rocko launcher fixtures (coordinator 48/48). Independently reproduce all D1-D10 and hostile controls. Especially review descriptor traversal: Linux/procfs only, all ancestor symlinks refused, open dir FDs anchor later reads, no-follow metadata open plus inode/device/mode checks, bounded reads, descriptor cleanup. Reject unintended external access or misleading portability guarantees. No live private roots, credentials, model calls, timers, Docker/build/release commands or source edits. Do not rerun host-wide suites with undisclosed effects. No shared-log writes while review is pinned. + +Review all-harness nested schemas and numeric fixture conversion, default enrollment/type membership, missing root/required structure/error distinctions, no partial list/library output and no input/key/value/parser excerpt diagnostics. Confirm credential.json siblings remain unopened and no unsafe fallback. The earlier 14/15 failing control is disclosed in progress note; it was diagnostic-order expectation, corrected before 43/43 and 48/48 passes. Other historical task-suite failures are not claimed fixed. + +Write ONLY docs/plans/reviews/2026-09-10_m20-correction-verdict.md: APPROVED or REQUEST CHANGES for exact correction, tested observations versus untested assumptions, any missing applicable regression gate. Send reply to Darkwing/default. Approval permits only this corrective source through selective commit/push; no materialization or live activation follows automatically. diff --git a/docs/plans/reviews/2026-09-10_m20-i2-correction-approval.md b/docs/plans/reviews/2026-09-10_m20-i2-correction-approval.md new file mode 100644 index 000000000..465bac894 --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-i2-correction-approval.md @@ -0,0 +1,9 @@ +# #1500 prerequisite correction approval gate + +M20-I2-CHARTER-FILBERT-1 charter APPROVED, implementation not authorized. Filbert freshly reproduced D1-D10 on clean b3fa2210 using synthetic fixtures: missing/empty roots accepted, out-of-root symlinks followed, incorrect version validation, nested unknown keys accepted, unenrolled defaults and incompatible credential types accepted, non-pi endpoints unchecked, URL credentials accepted, JSON parse excerpts leaked and partial list output emitted for invalid trees. + +These are defects in increment 1, not merely future hardening. The author and earlier review missed them; fifteen passing tests were inadequate evidence for the broad safety claim. No evidence here establishes real credential exposure; reproductions used synthetic data only. + +Request to Jason: authorize the narrow prerequisite correction in packages/mosaic source/tests/fixtures and its tracking records, through clean-tree regression tests, independent exact-candidate review and selective commit/push. Scope is the seven correction items in docs/plans/reviews/2026-09-10_m20-i2-charter-verdict.md. No materialization, refresh service, live credentials, package upgrades, deployment or unrelated work. No downstream implementation starts until correction verified. + +State: awaiting explicit owner implementation approval. Darkwing owns correction; Filbert independent verification with fresh adversarial cases. Existing charter remains draft for implementation authority. Reviewer acknowledgement dispatched to Filbert/default, transport exit 0/application acceptance unknown. No reply requested; next actionable event is Jason's approval or scope revision. diff --git a/docs/plans/reviews/2026-09-10_m20-i2-implementation-progress.md b/docs/plans/reviews/2026-09-10_m20-i2-implementation-progress.md new file mode 100644 index 000000000..e8341f973 --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-i2-implementation-progress.md @@ -0,0 +1,17 @@ +# #1500 fixture-only implementation progress + +Jason approved fixture-only materialization/refresh after correction publication at 27e4873a. Darkwing owns code and integration; Filbert final independent review. Rocko owns read-only PI 0.85.1 isolation research at M20-I2-ISOLATION-ROCKO-1. Both request and addendum dispatched through agent-send; initial dispatch exit0, application acceptance unknown. Rocko acknowledged original extraction was cleaned and a public exact-version re-fetch is underway; coordinator accepted temporary extraction only with identity verification and no scripts/CLI/credential use. Required return: declared isolation findings and public-source snapshot. No actual refresh driver until isolation is established. + +Implemented stages 2/3 locally: pure fixture execution resolution; explicit scope; enrolled defaults/overrides and immutable copied fork pins; no account fallback; unsupported native-model restriction refuses rather than falsely enforcing it. New credential store is in-memory and accepts only FIXTURE_ marker values, not arbitrary credential files. Workspace API creates a dedicated /tmp-only root; no caller-selected root or production backend. Exclusive retained execution claims, distinct generation paths, complete-file readback before publication, private modes, no credential digests in manifests, fixed failure/uncertainty receipts, and no same-ID blind retry. No service/CLI integration. + +Initial twelve materialization tests passed: concurrent different-account workspaces, fork rules, mutation isolation, private outputs, exclusive claims, injected partial-generation failures, lock contention, symlinked targets, and real-credential-shaped input rejection. Full generation readback validation added after that initial run and requires the next complete rerun. These are fixture observations, not production process or multi-record transaction guarantees. + +Next ready work: verify full candidate tests; implement fake-only refresh boundary when Rocko establishes exact target-version isolation; independently review complete snapshot before any source commit/push. Existing host-wide suites must not be invoked with undeclared Docker/state effects. Prior headless-suite identity failure is not waived; isolate and reconcile before claiming full regression completion. + +Stage23 review return: Filbert STAGE-SUPPORTED at snapshot9330fc5f, independently60/60 pass, not final approval. Added after-publish fault and new uncertainty regression per review note; materialization13/13 pass, log /tmp/m20-publication-uncertainty.log. Published generation survives uncertain outcome; no same-ID replay. New bytes require final whole-increment review. Reviewer acknowledgement exit0/application acceptance unknown. Rocko isolation findings remain the external prerequisite for refresh. + +Goal dispatch86baedd4: implemented standalone fixed fake refresh process in src/refresh-fixture.mjs, with clean explicit HOME/PI_CODING_AGENT_DIR/cwd, discarded child output, bounded timeout with child-close before cleanup, validated synthetic rotation output, no executable/environment injection. Focused tests7/7 pass at /tmp/m20-refresh-first.log. This is not yet integrated with store transactions/generation; next add integration and rotation/lock/failure regression tests, then freeze complete increment for Filbert. No real Pi invoked. + +Goal b3f597ab: integrated fixed fake refresh inside credential-store transaction. Near-expiry fixtures refresh automatically, options validate before claims, failures preserve prior generation/store state; same-account contention and independent-account success covered. Combined74/74 tests pass at /tmp/m20-refresh-integrated.log. Final complete candidate11255dd4 at /tmp/m20-refresh-final-review-xn34lpeb dispatched to Filbert as M20-I2-FINAL-FILBERT-1 via agent-send exit0/application acceptance unknown. Expected return docs/plans/reviews/2026-09-10_m20-refresh-final-verdict.md plus direct reply; Darkwing owns findings/integration. No publication yet. + +Final review return: M20-I2-FINAL-FILBERT-1 APPROVED11255dd4. Coordinator verified all21 reviewed bytes unchanged and clean baseline-plus-overlay74/74 pass, /tmp/m20-final-verified-overlay.log. Implementation/review goal complete, not publication completion. Remaining publication gate: reconcile broader charter suite applicability and historical headless failure without unapproved live test effects; then commit exact bytes and verify committed tree. No source changes after review. diff --git a/docs/plans/reviews/2026-09-10_m20-i2-prerequisites.md b/docs/plans/reviews/2026-09-10_m20-i2-prerequisites.md new file mode 100644 index 000000000..957344df2 --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-i2-prerequisites.md @@ -0,0 +1,7 @@ +# #1500 prerequisite measurement + +Baseline b3fa221060abb820cd809f15c3a54ba84518b7a1. Coordinator created an empty temporary directory and invoked the existing registry validate CLI with an absent child as --registry-root. Observed stdout `valid`, exit 0. Temporary parent removed; no real config or credentials read, no implementation edits. + +This reproduces a fail-open missing-registry defect, not just a source concern. The previous fifteen tests and independent approval did not cover it. Materialization must not depend on this validator until correction and independent negative-control verification. Other concerns in the increment-2 charter are hypotheses pending reproduction, not established defects. + +Next action: Filbert independently evaluates prerequisite findings and the charter; return a bounded corrective scope if needed. No materializer implementation or live service is authorized yet. The owner asked to draft the next charter and continue quickly, not to bypass failed prerequisites. diff --git a/docs/plans/reviews/2026-09-10_m20-i2-review-request.md b/docs/plans/reviews/2026-09-10_m20-i2-review-request.md new file mode 100644 index 000000000..f295a81dc --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-i2-review-request.md @@ -0,0 +1,7 @@ +# M20-I2-CHARTER-FILBERT-1 (#1500) + +Review draft at /tmp/m20-i2-charter-review-fga9lki3, two files, SHA256SUMS 78981e71521e5df2b62649fe11eda6c241029c4fe41ef5dfe06fff3776f44401; baseline b3fa221060abb820cd809f15c3a54ba84518b7a1. Author Darkwing, reviewer Filbert. Prior I1 reviewer role must be disclosed; fresh defect reproduction is required, not reliance on that approval. + +Owner requested drafting the next materializer/refresh charter and faster continuation, not implementation. Coordinator reproduced missing registry path returning valid/exit0. Independently reproduce in disposable git-archive tree and investigate charter prerequisite concerns using synthetic fixtures only. Do not read real credentials, launch Pi, arm timers, run Docker suites, install packages, change source or shared logs. No broad file scanning outside declared sources. + +Assess charter dependency ordering, exact scope, no-secret guarantees and rollback. Recommend a separate prerequisite correction if the existing validation defects prevent safely proceeding. Return APPROVED or REQUEST CHANGES for charter, with factual prerequisite findings distinctly separated from proposed design. Write ONLY docs/plans/reviews/2026-09-10_m20-i2-charter-verdict.md and reply to Darkwing/default. No implementation approval implied. diff --git a/docs/plans/reviews/2026-09-10_m20-publication-regression-request.md b/docs/plans/reviews/2026-09-10_m20-publication-regression-request.md new file mode 100644 index 000000000..bba89c76f --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-publication-regression-request.md @@ -0,0 +1,11 @@ +# M20-I2-PUBLICATION-GATES-FILBERT-1 + +Jason said "let's go" to regression reconciliation and publication. Reviewed source stays exactly11255dd4; Darkwing owns publication, you own independent regression applicability/reconciliation feedback only. No real credentials, Pi, Docker builds/releases, network model calls or live services authorized. + +Please inspect scripts/test-foundation.sh and scripts/test-conductor.sh and their callees before executing. If safe fixture-only, run unchanged scripts in disposable baseline27e4873a plus approved11255dd4 overlay with clean HOME/environment. Do not stash/edit live code or invoke stateful effects. Return precise results and any blocker. + +Also assess charter test-task/test-release applicability for this unintegrated fixture-only package. Do NOT waive a required gate. Those scripts contain real effects: release invokes activate; task contains real model/fork/recall tests, including unconditional recall at lines480+. Could offline evidence meet this slice's applicable checks, or does charter need explicit owner disposition before publication? Return recommendation with source evidence, not a blanket baseline waiver. + +Coordinator results: clean reviewed-overlay package+launcher74/74; config24/24 and auth15/15 in isolated HOME/env. Historical headless identity failure: compose.yaml26 forwards MOSAIC_AGENT_NAME; src/load-contracts.sh61 emits identity whenever set. Current host var is present. With no /var/lib/mosaic/user directory present, original unchanged loader was executed using three synthetic contract files and isolated env: identity set -> section present; unset -> absent. Two controls pass (/tmp/m20-headless-fixture.log). This proves contamination mechanism, not end-to-end reproduction of the historical suite run. Do not claim the full task suite now passes. + +Reply with artifact ONLY docs/plans/reviews/2026-09-10_m20-publication-regression-verdict.md and direct message. No source edits, commits/pushes or shared logs. Stop before any undeclared test effect. Production refresh remains excluded. diff --git a/docs/plans/reviews/2026-09-10_m20-refresh-final-review-request.md b/docs/plans/reviews/2026-09-10_m20-refresh-final-review-request.md new file mode 100644 index 000000000..5384b129e --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-refresh-final-review-request.md @@ -0,0 +1,17 @@ +# M20-I2-FINAL-FILBERT-1 + +Darkwing author; Filbert independent final whole-increment reviewer for owner-approved fixture-only #1500. No publication yet. + +Frozen candidate: /tmp/m20-refresh-final-review-xn34lpeb +SHA256SUMS digest: 11255dd4d76adb1762af0b5dd0d9eac114f48d5775594dd29ba554e4a96d82e6 +Contains complete packages/mosaic plus charter and corrected Rocko findings. Verify hashes before/after; test in a disposable writable copy only. Existing metadata implementation is baseline context. Earlier stage-supported verdict is not final approval. + +New stage: fixed fake subprocess uses isolated PI_CODING_AGENT_DIR/HOME/cwd with no inherited environment, no user executable injection, ignored stdio, timeout kill/child-close cleanup, bounded validated output and synthetic credential rotation. Generation runs it inside the store transaction on near-expiry OAuth credentials, or explicit refresh fixture option. Successful generation commits draft rotation. Failed refresh preserves prior generation and store state; same execution ID cannot retry. In-process lock contention refuses while independent account generation can proceed. + +Combined offline command: node --test packages/mosaic/tests/ scripts/test-darkwing-launch.mjs scripts/test-rocko-launch.mjs. Latest author result74/74 at /tmp/m20-refresh-integrated.log, plus syntax/diff checks. Snapshot includes package tests; use baseline launcher fixtures separately or copy unchanged launcher sources/tests into your disposable tree. Do not invoke broad shell suites that launch Docker, real Pi, or read live state. + +Review ALL increment files and test claims, not just refresh. Special attention: fixed-only runner restriction, marker-only data, output/error privacy, child cleanup/timeout, lock scope and rollback, same-ID tombstones, full-generation validation, fork semantics, and after-publish uncertainty regression added after early review. README explicitly notes in-memory locks, same-UID/hardlink residual and non-crash-atomic store/publication. Published uncertain generation remains but in-memory draft rolls back; fake-only simulation has no real token issuer. This is not a production refresh protocol. + +Rocko corrected findings hash6943eef5 confirms static 0.85.1 isolation. Model-catalog network suppression does not suppress OAuth refresh networking. Never execute real Pi, even with synthetic OAuth. No services/timers except bounded fake-child timeout, no real credentials/network/deployment. + +Return APPROVED or REQUEST CHANGES with exact findings and independent receipts in ONLY docs/plans/reviews/2026-09-10_m20-refresh-final-verdict.md, plus direct reply. Preserve earlier rejected/source findings and verdict history. No source writes, commits, pushes or shared session/build-log edits requested. diff --git a/docs/plans/reviews/2026-09-10_m20-refresh-isolation-request.md b/docs/plans/reviews/2026-09-10_m20-refresh-isolation-request.md new file mode 100644 index 000000000..713aac223 --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-refresh-isolation-request.md @@ -0,0 +1,9 @@ +# M20-I2-ISOLATION-ROCKO-1 (#1500) + +Jason approved fixture-only materialization/refresh implementation. Darkwing owns code; Rocko owns this read-only source investigation, Filbert reviews the final code independently. No live credentials, Pi launch, timers, services, model calls, installation or source edits. + +Using the already identity-verified 0.85.1 package from PI-REFRESH-ROCKO-1, determine exact supported environment/flags for `pi auth check --provider

` to use an isolated auth file, HOME and working directory with no native extension/config discovery. Read the actual CLI dispatch order: does auth-check load extensions/project packages/settings before handling the command? Is PI_CODING_AGENT_DIR or another override supported? What non-secret output and exit codes occur? Do not infer isolation merely from auth command naming. + +Copy only relevant public package docs/source plus package identity/checksum metadata to a dedicated /tmp/m20-pi0851-isolation-* snapshot accessible to coordinator/reviewer. No runtime/private files. If original extraction is unavailable, report that before fetching anew. Read complete relevant .md files and follow cross-references. No need to run the CLI, even with fake credentials. + +Write ONLY docs/plans/reviews/2026-09-10_m20-refresh-isolation-findings.md. Include exact source/file/line evidence and limits; flag inability to establish a safe isolated boundary as a blocker. Reply to Darkwing/default via agent-send.sh. No shared-log writes or additional task execution. diff --git a/docs/plans/reviews/2026-09-10_m20-stage23-review-request.md b/docs/plans/reviews/2026-09-10_m20-stage23-review-request.md new file mode 100644 index 000000000..9a60d68f4 --- /dev/null +++ b/docs/plans/reviews/2026-09-10_m20-stage23-review-request.md @@ -0,0 +1,11 @@ +# M20-I2-STAGE23-FILBERT-1 (#1500) + +Owner approved fixture-only stages after prerequisite correction at 27e4873a. Darkwing author, Filbert independent reviewer. Read-only early review of resolution and generation while Rocko investigates exact 0.85.1 refresh isolation; NOT final increment approval. + +Review these candidate files in the immutable snapshot specified below: src/execution.mjs, src/fixture-store.mjs, src/materialize-fixture.mjs, src/index.mjs and tests/materialize.test.mjs. Baseline records/registry code and existing fixtures are unchanged context. Copy the frozen packages/mosaic directory into a writable disposable temp copy for tests; verify source snapshot hashes before/after. Never run against live author files. There are no dependencies outside built-in Node for these tests. Run package tests there if useful. No Docker/live Pi/credentials/services, no shared-log edits, no implementation. + +Current combined package+launcher fixtures 60/60 pass; twelve new generation tests. Assess explicit scope as fixture inputs (not actual project membership authority), enrollment and fork pin rules, unsupported native-model ceiling refusal, in-memory-only store with synthetic marker language, exclusive execution claims retained on failure, atomic ready-directory publication and complete byte readback, fixed uncertainty/error records, no secret digests, no blind same-ID retry. APIs cannot accept a production credential adapter or output root; workspace is dedicated /tmp with cleanup. Source models shape based on installed docs; Rocko will verify 0.85.1 before final integration. + +Refresh is NOT implemented yet and must not be inferred. Find security/concurrency/claim/cleanup defects in this stage while source investigation proceeds. Return stage-supported or REQUEST CHANGES in ONLY docs/plans/reviews/2026-09-10_m20-stage23-verdict.md plus direct reply. Final whole-increment review will include refresh tests and target-version evidence. No commit/push implied by this early feedback. + +Frozen snapshot: /tmp/m20-stage23-review-51p59lff, SHA256SUMS 9330fc5f01ab046242ddef92dd11a21030729596640535b8f76927b38d6d825d, 17 files including unchanged context. Only the five named paths differ from baseline.