Files
stack/docs/remediation/BOARD.md
T
e80cebe7f0 docs(remediation): board reflects actual state + 4 dogfood findings from startup
Corrects the premature "DISPATCHED" planner line (per Mos): both planners are now
dispatched for real on GUARANTEED-clean context, not requested-clean.

Records Mos rulings: mission.json is retired-rail residue (do not invest); gate-16
holds on interim mos-dt-0 since rev-974 reviews; remote-control path is Mos-relay.

Captures 4 live failure classes observed while standing this seat up (D-1..D-4),
including an agent that silently ignored an in-message context reset — direct
evidence for the postmortem thesis that instructions are not enforcement.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
2026-07-31 17:09:58 -05:00

5.3 KiB

mos-remediation — LIVE BOARD (keep < 8 KB)

Phase: PLANNING — adversarial task decomposition IN FLIGHT. Updated: 2026-07-31 (mos-remediation orchestrator; seat active on mosaic-fleet).

Head

  • Mission charter + 15 decisions + 4-build plan: PERSISTED (docs/remediation/MISSION.md).
  • HOLD lifted for this workstream (Jason 2026-07-31). Nothing implemented yet — planning first.
  • Orchestrator seat mos-remediation is LIVE and owns the mission. Residency attestation: PASS.
  • TASK-0 (env + push mission package) IN PROGRESS — checkout deps repaired, gates being verified honestly.
  • TASK-1 (adversarial decomp) DISPATCHED for real, both planners on GUARANTEED-CLEAN context.

In-flight

Task Owner State
TASK-0 env repair + push remediation/mission-setup mos-remediation IN PROGRESS — deps installed; gates verifying
Decomp (robustness side) → DECOMP-OPUS.md planner-opus WORKING (clean session)
Decomp (pragmatic side) → DECOMP-SOL.md planner-sol WORKING (reset to 0.0% ctx before dispatch)
Reconcile both decomps → docs/remediation/TASKS.md mos-remediation BLOCKED on the two decomps

Fleet seats

  • mos-remediation — project orchestrator (Claude, /src/mosaic-stack, socket mosaic-fleet) — ACTIVE
  • planner-opus — adversarial planner (robustness), Opus 5, socket default — WORKING
  • planner-sol — adversarial planner (pragmatic), gpt-5.6-sol, socket default — WORKING
  • rev-974 — mosaicstack reviewer identity (id 16, write:repository) — idle, on call
  • Mos (mos-claude) — lead coordinator, socket default — relay path to Jason

Gate status

  • Delivery gates active: author≠reviewer, diff-blind pre-registered checks, CI-green, merged-PR completion.
  • Freeze: LIFTED for this workstream only.
  • Git identity: MOSAIC_GIT_IDENTITY=mos-dt-0 INTERIM. Mos ruled gate-16 HOLDS (author≠reviewer is what gate-16 protects; rev-974 reviews, mos-dt-0 never self-reviews). Dedicated identity TRACKED, Mos provisions.
  • Remote control: native /remote-control NOT wired in this runtime. Path is Mos-relay (Jason ↔ mos-claude via Discord ↔ mos-remediation via agent-send). Not a blocker.

Sequencing (from MISSION.md)

  1. Spine + choke-point service (MACP wiring @ mosaic_orchestrator.py::run_single_task) + PG/Redis
  2. Rotation daemon (finish Mission Control Plane, reuse packages/coord)
  3. Comms service (envelope→service→PG/Redis→adapters)
  4. Hygiene + conformance harness Cross-cutting retirements: flat-file tracking, 3 MACP islands, silent MOSAIC BYPASS.

Dogfood evidence captured this session (live failure classes, not hypotheticals)

  • D-1 / P-ACTIVATION + hygiene — committed .npmrc hard-pins store-dir=/root/.local/share/pnpm/store. Correct for the CI container (runs as root), fatal for EVERY non-root local checkout: EACCES on /root/.local/share/pnpm/store/v10/server/server.json. A committed config that only works on one runtime is exactly the activation-skew class. Fix candidate: make store-dir env-overridable, not hardcoded.
  • D-2 / hygiene — husky prepare fails EPERM copying into root-owned .husky/_/. Repo working tree has root-owned dirs (.husky/, repo root) under a non-root agent. Worked around with the intended HUSKY=0 escape hatch (does NOT disable the existing pre-commit/pre-push hooks).
  • D-3 / P-FLEET-001 — the seats running this mission are UNMANAGED. mos-remediation, rev-974, planner-opus, planner-sol appear in NO roster (~/.config/mosaic/fleet/roster.yaml, agents/). Planners run on socket default; the roster declares mosaic-fleet. This is the exact "one roster-owned socket/host + quarantine unmanaged + stale GC" failure P-FLEET-001 indicts — observed on the remediation mission's own fleet. Prerequisite for INBOX identity-addressing.
  • D-4 / P-LIFECYCLE + hygiene — a dispatched agent silently IGNORED an in-message context reset. planner-sol was at 64.3%/372k; the brief asked it to reset first; it began work on dirty context anyway. Only an out-of-band /new driven by the orchestrator guaranteed clean state. Confirms the postmortem thesis: instructions are not enforcement. Reset must be a mechanical pre-dispatch step, not a request.

Decisions log

  • 2026-07-31 — Mission set up by Mos post-postmortem (15/15 decided). Dogfood posture active.
  • 2026-07-31 — Mos: stale .mosaic/orchestrator/mission.json is RESIDUE of the disabled Python orchestrator rail that this plan RETIRES. Do NOT invest in it; do NOT build on that rail. The 0/0 milestone banner is cosmetic. (Supersedes any plan to repair it.)
  • 2026-07-31 — Mos: planners must be dispatched with GUARANTEED clean context, not requested-clean. Prior default-socket planner sessions predate this mission; dirty context is the indicted hygiene.
  • 2026-07-31 — mos-remediation: worker briefs forbid all git ops and restrict each worker to a single named output file, so two planners can share one checkout without a branch race (M2-era incident doctrine).