pdurlej/platform

Fork 0

checkpoint templates #636

Merged

pdurlej merged 4 commits from claude/m07/task-packet-templates into main

2026-06-01 08:47:27 +02:00

claude commented

2026-05-30 13:37:22 +02:00

Collaborator

Canary status: docs-only — schema spec + templates. No runtime mutation.

Canary Context Pack

Product story

The cross-cousin flow loop (#17) loses state at handoffs — Codex "gets lost mid-task" (the operator's observed ~1.5h self-correction loop), and claude rebuilds Codex's context from transcript pastes. ADR-0025 (Memory Control Plane) named the fix: explicit task memory. This PR lands the #460 keystone — the schema + templates that give Codex a tight envelope and let claude reconstruct state without a paste.

What changed (3 files in `state/memory/`)

CODEX_TASK_PACKET.md — the tight envelope claude hands Codex before runtime/CI work: objective, canonical sources, non-goals, machine-checkable done-criteria, checkpoint obligations, and a rescue protocol with a hard "stop after 2 self-corrections, write a checkpoint, hand back" rule.
CODEX_CHECKPOINT.md — observable-state snapshot template (checkpoint_kind enum) so claude resumes from the latest checkpoint alone — no transcript paste.
task-run-schema.md — the task_run / task_checkpoint schema spec + the metric field set (Opik-style). This is the contract #195's apply no-op counter emits against — same field names, no parallel metric system.

Why

ADR-0025 packet→execute pattern, operationalized. Protects Codex's budget (tight envelopes = less drift) and unblocks #195 (gives it the metric contract).

Evidence

Docs-only. No runtime mutation. The Postgres tables themselves land in a separate runtime PR (Codex lane, gated) — this is the schema spec they implement against.

Success criterion (measurable)

claude reconstructs Codex's state from the latest task_end/blocked checkpoint without a transcript paste.

Out of scope

Postgres table creation + platformctl wiring (runtime, Codex, gated).
Procedural registry (#461) — references procedures by slug in the packet.

Tracks #460. Provides the metric contract for #195. Feeds #461. Per ADR-0025.

For Codex: state/memory/task-run-schema.md § Metrics is the #195 contract — emit { task_id, repo, agent, event_kind: "apply_noop", count, ts } against those exact names.

Canary status: docs-only — schema spec + templates. No runtime mutation. ## Canary Context Pack ### Product story The cross-cousin flow loop (#17) loses state at handoffs — Codex "gets lost mid-task" (the operator's observed ~1.5h self-correction loop), and claude rebuilds Codex's context from transcript pastes. ADR-0025 (Memory Control Plane) named the fix: explicit task memory. This PR lands the **#460 keystone** — the schema + templates that give Codex a tight envelope and let claude reconstruct state without a paste. ### What changed (3 files in `state/memory/`) - **CODEX_TASK_PACKET.md** — the tight envelope claude hands Codex before runtime/CI work: objective, canonical sources, non-goals, machine-checkable done-criteria, checkpoint obligations, and a rescue protocol with a hard *"stop after 2 self-corrections, write a checkpoint, hand back"* rule. - **CODEX_CHECKPOINT.md** — observable-state snapshot template (`checkpoint_kind` enum) so claude resumes from the latest checkpoint **alone** — no transcript paste. - **task-run-schema.md** — the `task_run` / `task_checkpoint` schema spec + the metric field set (Opik-style). This is the contract **#195's apply no-op counter emits against** — same field names, no parallel metric system. ### Why ADR-0025 packet→execute pattern, operationalized. Protects Codex's budget (tight envelopes = less drift) and unblocks #195 (gives it the metric contract). ### Evidence Docs-only. No runtime mutation. The Postgres tables themselves land in a separate runtime PR (Codex lane, gated) — this is the schema spec they implement against. ### Success criterion (measurable) claude reconstructs Codex's state from the latest `task_end`/`blocked` checkpoint **without a transcript paste**. ### Out of scope - Postgres table creation + `platformctl` wiring (runtime, Codex, gated). - Procedural registry (#461) — references procedures by slug in the packet. Tracks #460. Provides the metric contract for #195. Feeds #461. Per ADR-0025. **For Codex:** `state/memory/task-run-schema.md` § Metrics is the #195 contract — emit `{ task_id, repo, agent, event_kind: "apply_noop", count, ts }` against those exact names.

claude added this to the 07 - Honcho memory v2 milestone

2026-05-30 13:37:22 +02:00

claude added the

agent/claude-code

label

2026-05-30 13:37:22 +02:00

claude added 3 commits

2026-05-30 13:37:22 +02:00

feat(memory): #460 CODEX_TASK_PACKET template 3f5b703ef3

feat(memory): #460 CODEX_CHECKPOINT template 576000d662

feat(memory): #460 task_run/task_checkpoint schema spec

canary-required / collect-diff (pull_request) Successful in 5s

Details

patchwarden-client-dry-run / collect-diff (pull_request) Successful in 6s

Details

base-is-main / guard (pull_request) Successful in 1s

Details

canary-required / canary (pull_request) Has been skipped

Details

patchwarden-client-dry-run / dry-run (pull_request) Failing after 1m49s

Details

patchwarden-pr-sanity / sanity (pull_request) Successful in 1m56s

Details

patchwarden-pr-sanity / collect-diff (pull_request) Successful in 7s

Details

2060f8ddb1

codex commented

2026-05-30 13:48:45 +02:00

Collaborator

Patchwarden PR sanity

Status: advisory_findings
PR: 636
Commit: cc933e5d976fd9faba5d36ca036434ca3059723b
Security-sensitive label: missing
Authority: advisory model review plus deterministic blockers only
3+3 canary: still alive; this does not replace it

Deterministic findings

No deterministic findings.

Model reviewers

`global-glm` / `glm-5.1:cloud`

Status: ok
Verdict: NOT_OK
blocker Empty diff - cannot verify claimed changes
- Evidence: Diff section is completely empty despite PR claiming 3 files changed in state/memory/ (CODEX_TASK_PACKET.md, CODEX_CHECKPOINT.md, task-run-schema.md)
- Next: Re-generate or re-fetch the diff. Cannot review schema correctness, metric field alignment with #195, or checkpoint_kind enum without seeing actual file content.

`global-deepseek` / `deepseek-v4-pro:cloud`

Status: ok
Verdict: OK
Findings: none

`redteam` / `kimi-k2.6:cloud`

Status: ok
Verdict: OK
Findings: none

Policy notes

GLM 5.1 + DeepSeek V4 Pro are the operator-required model mix for this bot.
Optional red-team model is enabled only when PLATFORMCTL_PR_SANITY_REDTEAM_MODEL is configured.
Auto-merge is not enabled here.

# Patchwarden PR sanity - Status: `advisory_findings` - PR: `636` - Commit: `cc933e5d976fd9faba5d36ca036434ca3059723b` - Security-sensitive label: `missing` - Authority: advisory model review plus deterministic blockers only - 3+3 canary: still alive; this does not replace it ## Deterministic findings No deterministic findings. ## Model reviewers ### `global-glm` / `glm-5.1:cloud` - Status: `ok` - Verdict: `NOT_OK` - **`blocker`** Empty diff - cannot verify claimed changes - Evidence: `Diff section is completely empty despite PR claiming 3 files changed in state/memory/ (CODEX_TASK_PACKET.md, CODEX_CHECKPOINT.md, task-run-schema.md)` - Next: Re-generate or re-fetch the diff. Cannot review schema correctness, metric field alignment with #195, or checkpoint_kind enum without seeing actual file content. ### `global-deepseek` / `deepseek-v4-pro:cloud` - Status: `ok` - Verdict: `OK` - Findings: none ### `redteam` / `kimi-k2.6:cloud` - Status: `ok` - Verdict: `OK` - Findings: none ## Policy notes - GLM 5.1 + DeepSeek V4 Pro are the operator-required model mix for this bot. - Optional red-team model is enabled only when `PLATFORMCTL_PR_SANITY_REDTEAM_MODEL` is configured. - Auto-merge is not enabled here.

pdurlej added 1 commit

2026-06-01 08:47:13 +02:00