chore(prompts): Codex RS2000 close-out dispatch (P0 phase3 apply chain + P1 recovery batch + P2 prune) #186

Merged

pdurlej merged 1 commit from claude/orders/codex-rs2000-close-2026-05-12 into main

2026-05-12 00:21:21 +02:00

claude commented

2026-05-12 00:18:33 +02:00

Collaborator

Tier classification per ADR-0007: Lite — single new prompt file, ~221 LoC, no sacred paths, no schema/runtime, no breaking change. Add label tier/lite.

Why now

Operator's 2026-05-12 ask: "zamknąć migracje RS2000 ASAP". Codex wakes ~01:00 CEST tonight; this prompt is the dispatch he reads cold.

What ships (1 file, +221 LoC)

prompts/codex-rs2000-close-2026-05-12.md — strict-order work queue:

P0 — Phase 3 apply chain close-out (#162→#167+#161): iterate ralph BLOCKER comments from 2026-05-11 00:54-00:59 CEST. Stacked-PR base-reset discipline included (modelizm-#4 avoidance). #161 needs base reset to main (stale). #167 has possible false-positive flagged for verify-first.
P1 — Recovery section batch (71 modules with TODO Phase 02): execute existing prompts/codex-recovery-batch-2026-05-11.md as drafted. Lite tier; ~14-15 PRs of 5 modules each. Count drift from 65 → 71 noted (new modules added).
P2 — RS2000 image prune (159 dangling, optional if budget allows): Codex drafts prompt; operator runs SSH manually per ADR-0003.

Operator flag inside the prompt

PR #168 (drafted by another claude thread, likely Prof Kong) uses ADR numbers 0006 + 0007 — conflicts with merged ADRs in PR #184. Codex is instructed to surface this to operator and NOT act unilaterally. Renumber or close-as-duplicate is operator-only decision.

What's NOT in this PR

Does NOT touch sacred paths, schemas, runtime
Does NOT dispatch Codex (operator pastes the prompt content into Codex app)
Does NOT execute any of P0/P1/P2 — just the dispatch artifact
Does NOT resolve the #168 ADR conflict (operator-only)

Test plan

Operator readback: does the priority order match operator's intent?
Operator readback: are the out-of-scope items correctly excluded?
Operator readback: is the #168 ADR conflict flag accurate?
Operator merge or operator-override per ADR-0001 Rule 2
Operator pastes prompt body into Codex app post-merge (or directly from branch — file is self-contained)

Spec sources read

state/STATUS_NOW.md (canonical operator snapshot)
prompts/codex-recovery-batch-2026-05-11.md (P1 plan referenced)
runbooks/recovery-section-template.md (recovery template)
decisions/0007-risk-proportional-canary.md (tier discipline)
Live Forgejo state: 81 modules cataloged v2 ✅; 71 with empty recovery; PRs #162-#167+#161 open with ralph BLOCKERs
Live RS2000 state: 60 containers healthy, 159 dangling images

Operator's North Star check

Does this PR reduce operator-attention-cost?

Reduces: Codex gets self-contained dispatch with strict ordering, sacred-path flags, ADR conflict surfaced. Operator's only job is paste + merge subsequent Codex PRs as they land.
Slight increase: operator decides #168 ADR conflict (one-time, ~5 min)

Verdict: ship.

🍵 — Drafted end-of-context-budget 2026-05-12 morning by claude (Pan Herbatka). Operator merge welcome anytime before Codex wakes ~01:00 CEST.

Refs: PR #184 (DeepSeek-findings remediation, merged), PR #185 (ADR-0016, merged), open PRs #161-#167 (Phase 3 apply chain), #168 (ADR conflict flag)

Tier classification per ADR-0007: **Lite** — single new prompt file, ~221 LoC, no sacred paths, no schema/runtime, no breaking change. Add label `tier/lite`. ## Why now Operator's 2026-05-12 ask: *"zamknąć migracje RS2000 ASAP"*. Codex wakes ~01:00 CEST tonight; this prompt is the dispatch he reads cold. ## What ships (1 file, +221 LoC) `prompts/codex-rs2000-close-2026-05-12.md` — strict-order work queue: - **P0 — Phase 3 apply chain close-out** (#162→#167+#161): iterate ralph BLOCKER comments from 2026-05-11 00:54-00:59 CEST. Stacked-PR base-reset discipline included (modelizm-#4 avoidance). #161 needs base reset to main (stale). #167 has possible false-positive flagged for verify-first. - **P1 — Recovery section batch** (71 modules with `TODO Phase 02`): execute existing `prompts/codex-recovery-batch-2026-05-11.md` as drafted. Lite tier; ~14-15 PRs of 5 modules each. Count drift from 65 → 71 noted (new modules added). - **P2 — RS2000 image prune** (159 dangling, optional if budget allows): Codex drafts prompt; operator runs SSH manually per ADR-0003. ## Operator flag inside the prompt **PR #168** (drafted by another claude thread, likely Prof Kong) uses ADR numbers **0006 + 0007** — **conflicts with merged ADRs in PR #184**. Codex is instructed to surface this to operator and NOT act unilaterally. Renumber or close-as-duplicate is operator-only decision. ## What's NOT in this PR - Does NOT touch sacred paths, schemas, runtime - Does NOT dispatch Codex (operator pastes the prompt content into Codex app) - Does NOT execute any of P0/P1/P2 — just the dispatch artifact - Does NOT resolve the #168 ADR conflict (operator-only) ## Test plan - [ ] Operator readback: does the priority order match operator's intent? - [ ] Operator readback: are the out-of-scope items correctly excluded? - [ ] Operator readback: is the #168 ADR conflict flag accurate? - [ ] Operator merge or operator-override per ADR-0001 Rule 2 - [ ] Operator pastes prompt body into Codex app post-merge (or directly from branch — file is self-contained) ## Spec sources read - `state/STATUS_NOW.md` (canonical operator snapshot) - `prompts/codex-recovery-batch-2026-05-11.md` (P1 plan referenced) - `runbooks/recovery-section-template.md` (recovery template) - `decisions/0007-risk-proportional-canary.md` (tier discipline) - Live Forgejo state: 81 modules cataloged v2 ✅; 71 with empty recovery; PRs #162-#167+#161 open with ralph BLOCKERs - Live RS2000 state: 60 containers healthy, 159 dangling images ## Operator's North Star check Does this PR reduce operator-attention-cost? - **Reduces**: Codex gets self-contained dispatch with strict ordering, sacred-path flags, ADR conflict surfaced. Operator's only job is paste + merge subsequent Codex PRs as they land. - **Slight increase**: operator decides #168 ADR conflict (one-time, ~5 min) Verdict: ship. 🍵 — Drafted end-of-context-budget 2026-05-12 morning by claude (Pan Herbatka). Operator merge welcome anytime before Codex wakes ~01:00 CEST. Refs: PR #184 (DeepSeek-findings remediation, merged), PR #185 (ADR-0016, merged), open PRs #161-#167 (Phase 3 apply chain), #168 (ADR conflict flag)

claude added 1 commit

2026-05-12 00:18:33 +02:00

chore(prompts): Codex RS2000 migration close-out dispatch for 2026-05-12

canary-required / collect-diff (pull_request) Successful in 4s

Details

canary-required / canary (pull_request) Successful in 11s

Details

6869db4e47

Strict-order work queue for Codex GPT-5.5 (sibling cousin):

P0 — Phase 3 apply chain close-out (#162→#167+#161)
  - Iterate ralph BLOCKER comments from 2026-05-11 00:54-00:59 CEST
  - #161 base needs reset to main (stale)
  - #167 has possible false-positive flagged for verify-first

P1 — Recovery section batch (71 modules with TODO Phase 02)
  - Execute prompts/codex-recovery-batch-2026-05-11.md as drafted
  - Lite tier per ADR-0007; ~14-15 PRs of 5 modules each

P2 — RS2000 image prune (159 dangling, optional if budget allows)
  - Codex drafts prompt; operator runs SSH manually per ADR-0003

Plus: PR #168 ADR-0006/0007 conflict flag for operator
(another claude thread drafted conflicting ADR numbers — operator decides
renumber or close as duplicate, NOT codex)

Out of scope tonight: vault→infisical, mail infra sunset, Phase 4 obs,
new ADRs beyond 0016, sacred-path changes without Full tier + operator
approval.

Reporting format + tradition (cousin letter on context close) included.

Refs: ADR-0006, ADR-0007, ADR-0016, REVIEW.md, codex-recovery-batch-2026-05-11.md