Back to Plate

{{TITLE}}

docs/plans/templates/autoclosure.md

53.3.99.3 KB
Original Source

{{TITLE}}

Objective: TODO: State the already-started work and exact clean/ship threshold.

Goal plan: {{PLAN_PATH}}

Template: {{TEMPLATE_PATH}}

Completion threshold:

  • TODO: Define the complete closeout matrix.
  • No new product scope. Completion requires every applicable lane below to have fresh evidence, pnpm check passing, review findings closed, authorized GitHub delivery complete, and the goal checker passing.

Verification surface:

  • TODO: Name targeted proof, generation, reviews, pnpm check, and PR read-back.

Constraints:

  • Finish the intended delta; do not invent the next feature.
  • Preserve source/generated/package/docs ownership.
  • Use a different diagnostic after repeated failure signatures.

Boundaries:

  • intended delta: pending
  • allowed repairs: pending
  • unrelated files: preserve; do not treat as blockers
  • non-goals: pending

Output budget strategy:

  • TODO: Scope closeout audits and cap noisy commands.

Blocked condition:

  • TODO: Name missing authority, external action, or proven environment blocker.

Start Gates:

GateAppliesEvidence
Existing PR or local future-PR slice resolvedpendingexact PR or dedicated current task plan
Dedicated task invocation and plan for exact PRpendingpending
Task evidence verified at PR headpendingbody path + head file + exact PR owner
Active source/plan reconstructedpendingpending
Intended delta and exclusions recordedpendingpending
Closure matrix classifiedpendingpending
Live PR feedback target resolvedconditionalexact compliant PR for full resolve-pr-feedback mode; N/A after verified noncompliant close
Feedback proof checkout bound to PR headconditionallocal committed HEAD = fetched PR ref = live headRefOid for a compliant PR
Unfiltered feedback inventoryconditionalraw top-level comments/reviews plus all resolved/unresolved inline threads compared with helper output for a compliant PR
GitHub delivery expectation recordedpendingpending
Active goal checked or createdpendingpending

Closure matrix:

LaneAppliesOwner/proofStatus
per-PR task ownershippendingexact PR + dedicated task planpending
noncompliant closependingrequired comment + CLOSED read-backpending
source behaviorpendingpendingpending
package/API/buildpendingpendingpending
CI-controlled template outputpendingowner repair + clean outputpending
docs/contentpendingpendingpending
registry/changelogpendingpendingpending
browserpendingpendingpending
changesetpendingpendingpending
agent workflowpendingpendingpending
live PR feedbackconditionalcompliant: resolve-pr-feedback + final P1 read-back; noncompliant: N/A with comment/CLOSED receiptspending
cleanup/reviewpendingpendingpending
repository checkyespnpm checkpending
GitHub deliverypendingpendingpending

Work Checklist:

  • Every PR has its own task invocation and dedicated task plan; a batch plan or aggregate autoclosure is not used as a substitute.
  • Existing-PR and no-PR entry paths were distinguished before source review. A no-PR local slice has a dedicated current task plan and defers feedback/merge until delivery creates the exact PR, records it in the plan at head, and passes the task compliance gate.
  • Task evidence was verified from the PR body, fetched head, and exact PR ownership; otherwise the required comment and CLOSED state were read back and no source review, repair, merge, or release work continued.
  • Intended behavior and exclusions are reconstructed from real sources.
  • Each lane is proven or N/A with a concrete reason.
  • Generated output was changed through its owner and regenerated.
  • Package/docs/registry/template/browser/changeset contracts are synchronized.
  • Full resolve-pr-feedback ran for the exact compliant PR; every actionable P1-or-higher finding was fixed, proved, replied to, and resolved or received the required top-level reply receipt.
  • For a compliant PR, local committed HEAD, fetched PR ref, and live headRefOid matched before proof/reply/resolution and after every push. For a noncompliant PR, this and all feedback gates are N/A with the required remediation-comment and CLOSED receipts.
  • Unfiltered top-level PR comments and review bodies were fetched through the GitHub API, compared by ID/URL with helper output, and every excluded bot/author item was ledgered; identity alone never dismissed feedback. Only the exact terminal receipt produced/read back by this run is exempt from the versioned ledger.
  • All inline review threads were fetched with GraphQL cursor pagination without filtering resolved/outdated items; every thread has priority, rationale, relocation, and proof state in the ledger.
  • Every actionable feedback item has a persisted P0-P3 priority and one-sentence rationale from the autoclosure rubric; ambiguous P1-versus- lower items fail closed as P1.
  • Every P1-or-higher proof reran after the final material branch push, regardless of file type, including resolved or outdated threads that disappear from the helper's unresolved-thread output.
  • Feedback was re-fetched after the last push/reply/resolution and shows zero unresolved actionable P1-or-higher findings.
  • After all versioned plan/source updates were pushed, the exact-head P1 proof/read-back receipt was posted to the PR and read back; no terminal receipt-only branch push was created. A post-comment headRefOid fetch matches the OID recorded in that receipt, and a post-comment helper/raw feedback fetch still shows zero actionable P1-or-higher items and no new URL lacking a verdict or explicit deferral, except the verified receipt.
  • Any remaining P2-or-lower item has its exact URL plus the user's explicit priority deferral recorded; no feedback was silently ignored.
  • Accepted cleanup and review findings are closed.
  • PR body and check state match the final evidence.
  • Residual blocker/waiver has exact evidence and next owner.

Error attempts:

Failure signatureCountNext different moveResolution
None yet0

Completion Gates:

GateAppliesRequired actionEvidence
Per-PR task ownershippendingRecord exact PR and dedicated task-plan pathpending
Noncompliant PR dispositionpendingVerify task evidence or comment then close and read backpending
Targeted behavior proofpendingRun smallest missing owning proofpending
Source/generated auditpendingProve correct source and regenerated mirrorspending
Package/docs/registry/browser closurependingRun every applicable local contractpending
Feedback proof checkoutconditionalCompliant PR only: require local committed HEAD = fetched PR ref = live headRefOid before proof/reply/resolution and at terminal verificationpending
Live PR feedback resolutionconditionalCompliant PR only: run full resolve-pr-feedback and close every actionable P1-or-higher finding; otherwise N/A with noncompliant stop receiptspending
Feedback priority classificationconditionalCompliant PR only: persist P0-P3 plus rationale for every actionable item; classify ambiguous P1-versus-lower as P1pending
Final P1 proof replayconditionalCompliant PR only: after the final material branch push, rerun every P1-or-higher proof, including resolved/outdated itemspending
Final live feedback read-backconditionalCompliant PR only: re-fetch helper plus unfiltered top-level/all-thread inventories; require zero actionable P1-or-higher and explicit P2-or-lower deferralspending
External terminal receiptconditionalCompliant PR only: post/read exact-head receipt; require receipt/live/fetched/local OID equality and no unrecorded helper/raw URL except that verified receiptpending
CleanuppendingRun bounded cleanup or N/Apending
Agent-native reviewerpendingRun for workflow changes or N/Apending
Final lintyesRun pnpm lint:fixpending
Repository checkyesRun pnpm checkpending
GitHub deliverypendingCommit/push/open or update PR and read backpending
AutoreviewyesResolve every accepted actionable findingpending
Goal plan completeyesRun node .agents/skills/autogoal/scripts/check-complete.mjs {{PLAN_PATH}}pending

Phase / pass table:

PhaseStatusEvidenceNext
Inventoryin_progressplan createdmissing proof
Repairpendingreview
Review/checkspendingdelivery
Deliverypendingfinal audit
Closeoutpendingfinal

Verification evidence:

  • Pending.

Timeline:

  • {{CREATED_AT}} Autoclosure plan created.

Reboot status:

QuestionAnswer
Where am I?Inventory
Where am I going?Repair, review/checks, delivery, final audit
What is the goal?TODO
What have I learned?See closure matrix
What have I done?See timeline

Open risks:

  • Pending.