stack/docs/plans/CURRENT.md

# CURRENT — single source of "what happens next"

This file always names exactly one next action. Any "continue" / "next" /
"proceed" message means: execute the action below, fully (implement → test →
verify against its acceptance criteria → commit → push → close the issue →
update this file to the next action). No ambiguity, no re-planning.

## Next action

Execute Jason's approved publication and planned-restart recovery trial (#1497).
Plan: `docs/plans/2026-09-08_publication-recovery-trial.md`. Publish remaining
reviewed changes, temporarily including agent definitions and labeled drafts;
exclude private runtime/session/auth state. Jason performed Rocko's planned restart
and issued neutral continue; PUB-REC-ROCKO-1 returned its reserved test PASS.
Filbert independently SUPPORTS bounded checkpoint recovery and APPROVES the pinned
privacy-safe summary (0e17757c); raw records stay local. No reliability or overall
trial acceptance inferred. WUI exact 53-file draft publication is independently
APPROVED with pending-acceptance label; source/default revisions remain unimplemented.
PUB-REC-FILBERT-LAUNCHER-1 APPROVED 23 source/publication candidates at
frozen manifest 8900faf1. Jason then changed Rocko's launcher to Sonnet.
PUB-REC-FILBERT-SONNET-2 APPROVED the six revised candidates at c5ad6306;
coordinator verified all seven R2 snapshot/live files unchanged. Remaining original
candidate approvals stand. Independent combined fixtures passed 5/5. Preserve
R1 REQUEST CHANGES and the historical Fable recovery evidence unchanged.
Publication completed at 29c1defe29e5793022e1d3820b265bfc0f7f628a on refactor;
92 exact approved units committed, committed-tree fixtures 5/5 passed, push succeeded
and fresh origin/refactor identity matched. Final aggregation APPROVED at 3f8e765b.
Next action: present the bounded publication/recovery trial to Jason for acceptance.
Do not close #1497 or resume the registry queue before that acceptance.
Newer executive-update skill edits, local ms-agent-watch deletion and additional
untracked skills are outside the pinned trial candidate. Preserve them uncommitted;
exclude both executive-update files rather than publish unreviewed or stale bytes.
Verify the prospective publication tree separately; live native launch would refuse
because its required ms-agent-watch file is now locally absent. No live launch or
unrelated skill migration is part of this trial.
No further restart or re-briefing requested. C1 remains HELD.

Registry review alignment is deferred behind this newly prioritized owner trial.
No registry implementation, agent relocation, main/next merge or deployment authorized.

Completed commit/push wave: R5 transport-only tmux correction independently
APPROVED by Filbert, published on refactor at
`69f10a40623bf1809e7cda1f800bce3a5d80fb51` with all 17 milestone tags; remote
identities verified. Exit 0 means transport dispatched, application acceptance
unknown. Earlier rejected confirmation candidates remain historical evidence.
Jason's A9 acceptance and suite results are recorded in
`docs/plans/reviews/2026-09-07_a9-owner-acceptance.md`.
#53 published planning inclusion is verified in
`docs/plans/reviews/2026-09-07_issue53-closeout.md`; Jason retains issue closure.
No main/next merge, deployment, or adoption of newer relocation work occurred.

Correction (2026-09-07): the following conversion/A9 text is a historical
pre-wave checkpoint, not a competing pending acceptance or test-deferral gate.
Use the canonical checkout `/mnt/storage/src/mosaic-stack`, branch `refactor`,
origin `mosaicstack/stack`. New foundation is at root; `v1/` is the legacy archive.
The old `~/src/mosaic-stack-dev-test` path is only a compatibility symlink.

Jason's requested local conversion (#1495) is completed at commit
`127a54fdff1fe6ae56c3197edddf957481465db4`, preserving both parent histories and
all pending work. Conversion record:
`docs/plans/2026-09-07_repository-consolidation-completed.md`.
Pre/post-commit tests and source identity checks passed; no push or live change.
The index is clean. Existing uncommitted work was preserved, not blanket staged.
Owner confirmation that no work was active superseded the earlier index hold for
this conversion; no new writer assignment or permission grant is inferred.

Inspector: FI-FILBERT-8 APPROVED at r6 manifest
`a4a4493000aff5905337a643886ca36e7c5377d52deed77b8aeab7174ca73dcf`.
All 382 file identities and pins remain unchanged. Demo guide:
`docs/plans/reviews/2026-09-07_foundation-inspector-demo.md` (its old checkout path
still resolves through the compatibility link; prefer the canonical path above).
Four examples reran successfully after conversion. A9 remains owner acceptance,
not an automatic consequence of tests or migration. Task/release coverage remains
NOT RUN/DEFERRED, not deployment green. Native-parser/equality qualifications remain.

Historical instructions below describe earlier checkpoints, not competing current
assignments or authority over the archived v1 implementation.

## Earlier owner and source checkpoints

Historical context below; the current candidate and live goal checkpoint supersede
the earlier partial-draft descriptions.

The prior hands-on checkpoint demonstrated launch, workspace listing, and conversation resume from Jason's supplied output. Fresh context and mission recovery were not tested. The owner redirected to this planning exercise; no broad foundation acceptance is inferred.

Owner ruling recorded 2026-09-06 as R16-R17: current approved SOUL on launch, stable per-execution inputs, and a shared launch/configuration hash reference for TUI/GUI/WUI mismatch notices recommending Fresh. D10 is partly resolved. Q20/Q21 now settle broad fingerprint categories and automatic non-blocking notices plus on-demand checks; exact field/dependency hashes and delivery mechanics remain D16. This does not advance the phase or authorize implementation.

Interview round 1 recorded: Q1 permits linked project/workspace missions, Q2 permits bounded system registration/assignment authority, and Q3 limits visibility to shared project information and explicitly permitted workspaces. Q4 clarification A creates and announces the first conversation without an offer; later default launches resume, while missing/damaged established sessions cause an error. Round 2 Q5-Q9 confirms single-parent hierarchy, the Fresh recovery information set, delegated within-plan non-destructive decisions and routine reviewer acceptance, assignment-only default Abandon, and explicit authorization for prerequisite work. Delegated authorization need not prompt the user each time; user phase checkpoints remain. Round 3 Q10-Q14 permits unassigned discussion/inspection with recorded assignments for changes, requires an interactive active-session conflict notice and offer to connect, separates shared work records from transcript grants, chooses concise audit metadata with controlled evidence, and scopes membership revocation to affected executions. Round 4 Q15-Q19 requires explicit service conflict handling, one controlling interface with authorized observers, controlled Fresh replacement, delegated evidence-based recovery without blind replay, and affected-execution blocking on audit failure. Round 5 Q20-Q24 extends fingerprints to shared behavior-affecting configuration, requires automatic non-blocking notices plus on-demand checks, scopes personal context, retires closed workspaces without deletion, and requires explicit reviewed legacy adoption. Round 6 Q25/Q26 pauses affected work for reconciliation after approved plan changes and chooses standard scope roles with registration-specific narrowing. Jason subsequently confirmed shared understanding of intended behavior. Jason then authorized phase 2. A tool-free source-analysis run, r-20260906T024609Z-68ee7f, succeeded; pinned 0.84.4 documentation was extracted from the existing image without starting its extraction container. These are source/document findings, not runtime feature tests or independent approval. The first contract candidate is partial. Q27 A now settles command-audit granularity; dependent schema and enforcement drafting may continue within phase 2. Full schema/plan approval remains pending.

## Accepted phase-2 checkpoint (historical)

- Goal: issue-53-phase2. Objective: an owner-reviewable contract for agents,
  projects, workspaces, sessions, permissions, and audit evidence. Completion
  owner: Jason. Author/workspace/session remain those recorded below.
- State: satisfied. Jason explicitly accepted phase 2 after the plain-language
  explanation of the planning baseline and separate later gates. P2-7 is complete.
  REVIEW.md retains D1-D16 and the unproved implementation mechanisms. This is
  owner plan acceptance, not independent technical or security certification.
- Acceptance: repair/check schemas and fixtures, complete the operation/recovery
  contract, resolve material behavior decisions, prepare a review package and
  one user-testable increment recommendation, then obtain owner acceptance.
- Evidence: `python3 docs/plans/foundation-v1-candidate/check.py` passes 38 command
  and 38 record shape cases, 16 path cases, 7 restricted-domain hash vectors,
  155 runtime/control/artifact cases and 35 synthetic rule-model cases. Ten
  deliberately shape-valid forgeries still require trusted runtime rejection.
  These are not runtime security tests or independent acceptance.
- Reboot fixture defects were repaired, not discarded. Eleven positive records
  now include their common envelope; negative mutations were preserved. Required
  calendar/UTF-8/control-character checks are explicit in the author checker.
- Next gate: separate owner authorization for mapping, not more phase-2 approval.
  No mapping or implementation started. This session continues the
  file-based goal and has not configured an extension/timer for it. Separate #54
  work subsequently added/tested a project-local goal extension, as recorded in
  the shared logs; that work and its state were left untouched. This session has
  not migrated issue-53-phase2 into that runtime. Elapsed time never grants approval.
- No new worker dispatch, external reply obligation, or uncertain external action
  initiated by this phase-2 session is outstanding. No numeric work budget supplied;
  aggregate usage remains unavailable.
- Authority remains phase-2 planning and read-only investigation. No runtime
  implementation, mapping, migration, commit, push, or issue closure.

## Prior recovery checkpoint, 2026-09-06 03:42 UTC

Historical snapshot below; the live goal checkpoint above supersedes its pause
and unfinished-fixture status.

- Goal: issue-53-phase2. State: paused by owner steering. Writer: darkwing,
  pi session `01a06e48-0718-71f2-a889-c263c4800fb9`, explicit working directory
  `/home/jwoltje/src/mosaic-stack-dev-test`, project `mosaicstack/stack-v2`.
  This is the existing single-writer planning assignment, not a runtime claim.
- HEAD remains `69d1bb3`. Preserved all uncommitted planning and unrelated skill
  work. No reset, cleanup, commit, push, or implementation occurred.
- Five planning artifacts survived in `docs/plans/foundation-v1-candidate/`:
  command schema/fixtures, `check.py`, and record schema/fixtures. The three
  command-check file hashes match the pre-reboot checksums.
- `python3 docs/plans/foundation-v1-candidate/check.py` passes 38 shape fixtures
  and 5 deliberate shape-valid forgeries. This does not prove runtime security.
- The unfinished record checker is NOT integrated into check.py. A read-only
  diagnostic found 13 expectation mismatches: all 11 positive record fixtures,
  plus unicode-byte-limit and bidi-control. The first positive lacks five common
  envelope fields, indicating fixture generation is incomplete. Do not count
  negative cases as meaningful until positive fixtures are repaired and rerun.
- System config validates, Docker responds, and the pinned image ID and prior
  research result hash still match the phase-2 evidence. No Mosaic worker
  container was running at inspection. Pinned temporary docs remain available.
- Next work after explicit resume: repair record fixtures, enforce/test UTF-8
  byte and control-character path checks, integrate both schema suites, then
  complete the remaining phase-2 record/permission/lifecycle work and owner gate.
- No outstanding assistant-initiated external action or reply is known. Wake is
  manual: Jason sends a resume instruction. No timer or automatic continuation
  is registered. Aggregate usage is unavailable; no numeric budget was supplied.

## Queue (ordered per docs/plans/ROADMAP.md)

1. Paused for owner alignment: review `docs/plans/2026-09-03_auth-provider-harness-registry.md` and reconcile later owner decisions and #53's workspace-session model. Gate 7 remains unresolved. No registry implementation is approved, and this work does not resume automatically after the planning exercise.
2. Deferred by owner: CI runners (Gitea hardware slow); second real adapter; push automation

## Rules

- One action in flight. Update this file at the END of every action.
- Blocked? Move the item to "Blocked" below with the reason and stop.
- Completed actions move to the log at the bottom (date + issue + result).
- Corrected entries are marked, never silently rewritten (see 2026-09-03 dedup note).

## Blocked

(none)

## Completed log

- 2026-09-08 — Owner-requested development team: Rocko launches Claude Code with Fable; Filbert launches Pi with `openai-codex/gpt-6-astra:low`; Darkwing is the development team lead and Dewey remains frontend/UX owner. Five offline launcher tests and both new agents’ local configuration checks pass. No model session started; inspector acceptance queue unchanged.

- 2026-09-08 — Owner-requested Dewey frontend/UX agent: `agents/dewey/launch.sh` uses the shared native launcher with its own persona, context and session history. Offline launcher checks pass for both agents and real Dewey `--check` passes; no model session started. Current inspector acceptance queue unchanged.

- 2026-09-07 — Owner-directed concept annexation: [Mosaic concepts](../concepts/README.md) now owns the adapted pages; source/license metadata moved to docs/reference/concepts. Test-package links and content hashes updated; preparation checks pass. Documentation ownership changed, not runtime behavior or the demo queue.

- 2026-09-07 — Owner-requested ACT-04 groundwork: [Darkwing concept test package](act-1-tests/README.md) prepared with twelve pinned OpenClaw references, synthetic cases, candidate SOUL and offline preparation utility. Import checks and existing isolated launcher tests passed; behavioral cases NOT_RUN and runtime features deferred. Current demo queue unchanged.

- 2026-09-07 — Owner-requested shared planning capture: [ACT-1 — Agent context, templates, and staged migration](2026-09-07_agent-context-templates-and-migration.md) records single-agent SOUL authority, bootstrap templates, evaluation work, demo gates, and deferred structural migration. Runtime tasks remain unassigned; current demo queue unchanged.

- 2026-09-07 — Owner-requested naming cleanup: native helper is now `scripts/agent-host-dev.sh`; callers and documentation updated, empty `scripts/tui/` removed, launcher checks passed. Historical log paths retain their original names.

- 2026-09-07 — Owner-requested entry-point consolidation: `scripts/agent.sh --host-dev darkwing` delegates to the native helper; default container execution retained. Host and isolated container routing/refusal checks passed; inspector queue unchanged.

- 2026-09-07 — Owner-requested launcher follow-up: Darkwing's launch.sh now delegates to `scripts/tui/launch.sh darkwing`; existing session/context regression checks pass.

- 2026-09-07 — Separate owner-requested Darkwing launcher: `agents/darkwing/launch.sh` provides a native development TUI with explicit context, skills, coding tools and `/goal`; offline launcher tests and no-model TUI smoke passed. Inspector queue and approval gates unchanged; no commit/push.

Note (2026-09-03): this log was deduplicated after editor-session races
appended duplicate blocks. The dedup removed repeated lines only; every
distinct action appears exactly once, in completion order. Ground truth:
git history + Gitea issues.

- 2026-09-03 — POC: containerized pi hello-world (poc-container-hello-v0)
- 2026-09-03 — M1 configuration-driven hello world (#1–#4; config-hello-v1); hotfix #5 stdin detach
- 2026-09-03 — M2 mission/task abstraction (#6–#9; mission-task-v1); hotfix #14 release identity in task path
- 2026-09-03 — M3 release model + safe updates (#10–#13; release-model-v1); drills: update/refusal/rollback
- 2026-09-03 — M14 live user context layer (user/ dispatched to all launches; 0.0.9 built)
- 2026-09-03 — M15 agent seats: per-agent SOUL + role contracts (#36; agent-seats-v1); roles/ convention (root = bootstrap-only)
- 2026-09-03 — M13 interactive TUI agent + TOOLS.md (#35; interactive-agent-v1); release 0.0.8 activated
- 2026-09-03 — M12 conductor auto-apply policy (#34; auto-apply-v1); 17 conductor selftests
- 2026-09-03 — M11 session forking (#33; session-fork-v1); child recalls ancestor, base untouched
- 2026-09-03 — M10 run-record retention (#32; retention-v1); prune keep-N, dry-run default, receipt
- 2026-09-03 — M9 mission capability policy (#30; mission-policy-v1); least-privilege intersection
- 2026-09-03 — test UX: green OK/red FAIL status colors; NO_COLOR-aware
- 2026-09-03 — M10-era hotfix: retry lineage (#28) + AGENTS.md/SESSIONS.md recovery shim
- 2026-09-03 — release 0.0.10 packaged and health-gated activated (user context + agent seats live)
- 2026-09-03 — release 0.0.11 shipped (onboarding + live user context); ROADMAP.md agreed (M16–M19); CI deferred by owner
- 2026-09-03 — M16 release self-determination (#38; `release.sh ensure` at launch, drift warnings, recursion guard) — logged late: CURRENT.md had gone stale while M16/M17 shipped; ground truth = git history
- 2026-09-03 — M17 skill lifecycle + ms-* skill set completion (#40–#42; skill-lifecycle-v1); release 0.0.12 packaged, health-gated active — logged late, same staleness correction
- 2026-09-03 — conductor-loop calibration with live collaborator (#43): dispatch via agent-send.sh → receipt → line-by-line diff review → suite-gated integration; docs/TOOLS.md gains Tools (host-side) section + corrected suite counts
- 2026-09-03 — skill revisions adjudicated (#44): ms-communications integrated as-authored; ms-conductor redraft + conductor remediation (refusal vs outage); TOOLS.md release.sh ensure row
- 2026-09-03 — M18 seat-role progressive capability restriction (#45; roles resolve to contracts, ceiling ∩ seat grant, fail-closed refusals, roles/researcher.json); task suite 74 → 88
- 2026-09-03 — M18 follow-up: fail-closed seat resolution under MOSAIC_AGENTS_DIR override (#46, owner decision after live verification); task suite 88 → 90; next action M19
- 2026-09-03 — M19 harness auth tooling (#47; auth.sh status/accounts, agent.sh --auth per-launch injection via PI_AUTH_FILE, test-auth suite 13 cases with secret-never-printed assertions); agreed sequence M16–M19 complete, M20 owner-gated
- 2026-09-03 — M19 correction: auth ownership moved to the data root (#48, owner direction — the stack never writes to default harness config locations; ROADMAP standing decision); auth.sh config-driven, accounts at <dataRoot>/auth, 0600 enforced; test-auth 13 → 15
- 2026-09-03 — harness/provider/auth registry specification drafted (#49): agent.json harness declaration, central provider/account/settings registries, runtime seat selection, mechanical per-harness materialization, centralized OAuth refresh, Ollama endpoints, CLI contract; implementation blocked pending ten-gate review