goal repair mode

Objective: Add goal repair <expectation> workflow support so goal and derived skills can repair the owning rule/template after a completed or in-progress goal misses user expectations.

Goal plan: docs/plans/2026-05-25-goal-repair-mode.md

Template: docs/plans/templates/task.md

Task source:

• type: user prompt
• id / link: chat request
• title: add goal repair <expectation>
• acceptance criteria: goal documents repair mode, generated goal skill syncs, a lightweight goal-repair template exists, smoke generation/checker proof passes, and repair mode stays narrow enough to avoid broad self-editing.

Completion threshold:

• .agents/rules/goal.mdc documents repair <expectation> mode with target selection, repair classification, safety rules, and verification.
• .agents/skills/goal/SKILL.md is regenerated from the source rule.
• docs/plans/templates/goal-repair.md exists as the repair plan shell.
• A fresh repair smoke plan includes the repair rows and fails check-complete.mjs while unfinished.
• node .agents/rules/goal/scripts/check-complete.mjs docs/plans/2026-05-25-goal-repair-mode.md passes.

Verification surface:

• cwd plate-2: pnpm install
• cwd plate-2: rg source/generated/template audit for repair mode rows
• cwd plate-2: node .agents/rules/goal/scripts/create-goal-scratchpad.mjs --template goal-repair --title "repair smoke"
• cwd plate-2: unfinished repair smoke plan fails check-complete.mjs
• cwd plate-2: targeted Biome attempt for touched markdown/rule files
• cwd plate-2: final check-complete.mjs on this plan

Constraints:

• Keep repair mode narrow: one expectation, one owner, smallest durable repair.
• Patch source-of-truth files, not generated skill mirrors.
• Do not weaken evidence safety or completion gates to satisfy annoyance.
• Do not add broad mechanical enforcement unless prose/template repair is insufficient.

Boundaries:

• Source of truth: latest user prompt plus pasted goal, task, and slate-plan skills.
• Allowed edit scope: .agents/rules/goal.mdc, .agents/skills/goal/SKILL.md, docs/plans/templates/goal-repair.md, and this plan.
• Browser surface: N/A: no browser behavior.
• Tracker sync: N/A: no tracker item.
• Non-goals: no changes to task or slate-plan in this slice; goal repair targets their templates/rules when invoked later.

Blocked condition:

• Autonomous work stops if the generated goal skill cannot sync, the repair template cannot instantiate, or the final goal plan cannot pass check-complete.mjs after one targeted fix.

Task state:

• task_type: workflow / agent-tooling feature
• task_complexity: non-trivial, auditable
• current_phase: closeout
• current_phase_status: complete
• next_phase: final response
• goal_status: active until final check passes

Current verdict:

• verdict: complete after final plan check
• confidence: high
• next owner: none
• reason: source rule, generated skill, repair template, and smoke checks are aligned.

Completion rule:

• Do not call update_goal(status: complete) while any required checklist item remains unchecked. If an item does not apply, check it and add N/A: <reason>.
• Do not call update_goal(status: complete) until every completion threshold above is satisfied, final handoff evidence is recorded, and node .agents/rules/goal/scripts/check-complete.mjs docs/plans/2026-05-25-goal-repair-mode.md passes.
• Do not create hook state for this goal. This file plus the active goal are the durable state.

Start Gates:

Gate	Applies	Evidence
Skill analysis before edits	yes	Read pasted goal, task, and slate-plan; read local .agents/rules/goal.mdc, .agents/rules/task.mdc, and templates before editing.
Active goal checked or created	yes	get_goal returned none; create_goal created the repair-mode objective.
Source of truth read before edits	yes	User prompt and pasted skills read before patching.
Tracker comments and attachments read	N/A: no tracker	No tracker source.
Video transcript evidence required	N/A: no video	No video evidence.
docs/solutions checked for non-trivial existing-code work	N/A: workflow-template update	No product/code bug pattern lookup needed.
TDD decision before behavior change or bug fix	N/A: no runtime behavior bug	Template smoke and source audit are the correct proof.
Branch decision for code-changing task	N/A: no branch change requested	Continued on current checkout.
Release artifact decision	N/A: no package release	No package behavior or published API changed.
Browser tool decision for browser surface	N/A: no browser surface	No route/UI behavior changed.
PR expectation decision	N/A: no PR requested	No PR in this task.
Tracker sync expectation decision	N/A: no tracker	No tracker sync.

Work Checklist:

• Objective includes outcome, completion threshold, verification surface, constraints, boundaries, and blocked condition.
• Task source classified with source type, id/link, title, task type, acceptance criteria, caveats, likely files/routes/packages, browser surface, and root-cause layer.
• Required video or screen-recording evidence is cached/read as normalized <video-transcripts> XML, or marked N/A with reason.
• Nearby repo instructions and implementation patterns read before edits.
• Implementation fixes the right ownership boundary, or the narrower choice is recorded with reason.
• Release artifact requirement recorded: changeset, registry changelog, or N/A with reason.
• Final handoff shape decided: bug/feature/testing/batch/review/tracker requirements, PR body sync, and issue/Linear sync when applicable.
• Branch handling recorded for code-changing work: dedicated branch used, new branch needed, or N/A with reason.
• Local-env-rot retry policy recorded for any surprising repo-wide failure: reinstall/rerun evidence or N/A with reason.
• Workspace authority recorded: every proof command names the cwd/tool that owns the changed behavior.
• High-risk note recorded for public API, runtime, package-boundary, browser behavior, agent-action, or command-contract changes, or marked N/A with reason.
• Review/autoreview target selected from actual diff state for non-trivial implementation work, or marked N/A with reason.
• Agent-native review decision recorded for .agents/**, .claude/**, .codex/**, skills, hooks, commands, prompts, or user-action tooling.

Completion Gates:

Gate	Applies	Required action	Evidence
Named verification threshold	yes	Run the command, proof, source audit, or artifact check named in this plan	pnpm install, rg, repair smoke generation, incomplete-plan check, targeted Biome attempt, and final plan check recorded.
Bug reproduced before fix	N/A: no bug fix	Record failing test/repro or N/A with reason	Workflow feature, not a bug repro.
Targeted behavior verification	yes	Run focused test/proof for changed behavior or record N/A	Repair smoke plan included repair rows and failed check-complete.mjs while unfinished.
TypeScript or typed config changed	N/A: markdown/rule files only	Run relevant typecheck	No typed source/config changed.
Package exports or file layout changed	N/A: no package exports	Run pnpm brl before final verification and keep generated barrel updates	No package barrel/export changes.
Package manifests, lockfile, or install graph changed	N/A: no package graph target	Run pnpm install and relevant package checks	pnpm install ran for Skiller sync and completed.
Agent rules or skills changed	yes	Run pnpm install and verify generated skill sync	pnpm install completed; rg confirmed .agents/skills/goal/SKILL.md includes repair mode.
Workspace authority proof	yes	Run verification in the owning repo/package/app/route/tool and record cwd; do not count the wrong workspace as proof	All proof ran from cwd /Users/zbeyens/git/plate-2, which owns .agents and docs/plans/templates.
Browser surface changed	N/A: no browser surface	Capture Browser Use proof or record explicit waiver/blocker	No browser surface.
Browser final proof	N/A: no browser surface	Attach screenshot or exact browser verification caveat when browser proof applies	No browser proof required.
CI-controlled template output changed	N/A: no templates/** edit in this task	Restore generated template output or record why it is intentionally kept	Touched docs/plans/templates, not CI-controlled templates/**.
Package behavior or public API changed	N/A: no published package behavior	Add a changeset or record why no changeset applies	No changeset.
Registry-only component work changed	N/A: no registry component work	Update docs/components/changelog.mdx or record N/A	No registry work.
High-risk mini gate	yes	For public API/runtime/package-boundary/browser/agent-action/command-contract changes, record realistic failure mode, proof plan, and why the chosen boundary is right; otherwise N/A	Failure mode: repair becomes broad self-editing or weakens gates. Proof plan: one-expectation safety rules, target classification, repair template, source/generation audit, smoke incomplete-plan failure. Chosen boundary: generic repair command lives in goal; derived skills are repaired only when invoked/named.
Agent-native review for agent/tooling changes	yes	For .agents/**, .claude/**, .codex/**, skills, hooks, commands, prompts, or user-action tooling, load .agents/skills/agent-native-reviewer/SKILL.md and close accepted/actionable findings, or record N/A	Loaded agent-native reviewer. Manual incremental review: no user-only action is added; this exposes an agent workflow command and documents source/template ownership. No parity gap.
Local install corruption suspected	N/A: no corruption signal	Run pnpm run reinstall once, rerun the exact failing command, or record N/A	No install corruption signal.
Autoreview for non-trivial implementation changes	N/A: source-rule/template docs patch, no runtime implementation	Load .agents/skills/autoreview/SKILL.md; use dirty local --mode local, branch/PR --mode branch --base <base>, or committed slice --mode commit --commit <ref> until no accepted/actionable findings, or record N/A for docs-only/trivial/no local patch	No runtime implementation patch; agent-native/source smoke checks apply.
PR create or update	N/A: no PR requested	Run check before PR work and sync PR body to final handoff	No PR in this task.
PR proof image hosting	N/A: no PR/browser image	If PR body needs browser proof, replace local image paths with hosted GitHub URLs or record N/A	No PR image.
Tracker sync-back	N/A: no tracker	Post concise issue/Linear sync after PR exists, or record N/A/blocker	No tracker sync.
Final handoff contract	yes	Fill the final handoff fields below with exact PR/issue/confidence/tests/browser/outcome/caveats/design/verification content or N/A reason	Final handoff fields below are filled.
Final lint	N/A: touched markdown/rule paths ignored by Biome	Run pnpm lint:fix or scoped equivalent	pnpm exec biome check <touched files> --fix processed 0 files because paths are ignored by config.
Goal plan complete	yes	Run node .agents/rules/goal/scripts/check-complete.mjs docs/plans/2026-05-25-goal-repair-mode.md	Run after this final plan update.
Knowledge extraction	N/A: source rules now contain reusable knowledge	Evaluate ce-compound; capture if useful	No separate compound note needed.

Phase / pass table:

Phase	Status	Evidence	Next
Intake and source read	complete	Prompt, pasted skills, local rule/template reads	implementation
Implementation	complete	Patched goal rule and added goal-repair template; pnpm install regenerated skill mirror	verification
Verification	complete	Source audit, generated skill audit, repair smoke, incomplete-plan guard, Biome ignored-path note, agent-native review	closeout
PR / tracker sync	complete	N/A: no PR/tracker	final response
Closeout	complete	This plan ready for final check-complete.mjs	final response

Findings:

• goal had template creation guidance but no way to repair templates/rules after a completed goal failed a user expectation.
• Derived skills like task and slate-plan can stay unchanged because repair mode targets their owning rules/templates when invoked with a named skill or plan.

Decisions and tradeoffs:

• Add repair <expectation> to goal rather than each derived skill.
• Add a dedicated goal-repair template so repair work records expectation, observed miss, ownership, classification, proof, and non-repairs.
• Do not add mechanical checker logic yet. The first repair mode can be source/template-driven; scripts come later only if prose gates keep failing.

Implementation notes:

• Updated .agents/rules/goal.mdc description and argument hint.
• Added Repair Mode with trigger, use/non-use cases, target selection, repair classification, workflow, and safety rules.
• Added docs/plans/templates/goal-repair.md.
• Ran pnpm install; generated .agents/skills/goal/SKILL.md synced.

Review fixes:

• None.

Error attempts:

Error / failed attempt	Count	Next different move	Resolution
Targeted Biome check processed no files	1	Record as N/A for markdown/rule files	Biome config ignores touched paths.

Verification evidence:

• pnpm install completed and Skiller applied rules for Claude Code and Codex.
• rg found repair <expectation>, ## Repair Mode, target selection, repair classification, --template goal-repair, and safety rules in both .agents/rules/goal.mdc and .agents/skills/goal/SKILL.md.
• rg found the repair template expectation and completion rows in docs/plans/templates/goal-repair.md.
• node .agents/rules/goal/scripts/create-goal-scratchpad.mjs --template goal-repair --title "repair smoke" created a smoke plan.
• node .agents/rules/goal/scripts/check-complete.mjs docs/plans/2026-05-25-repair-smoke.md; test $? -ne 0 proved the unfinished smoke plan fails.
• Deleted the smoke plan after verification.
• pnpm exec biome check .agents/rules/goal.mdc .agents/skills/goal/SKILL.md docs/plans/templates/goal-repair.md docs/plans/2026-05-25-goal-repair-mode.md --fix processed 0 files because these paths are ignored.

Final handoff contract:

• PR line: N/A: no PR requested.
• Issue / tracker line: N/A: no tracker.
• Confidence line: high; repair mode is in source, generated skill is synced, and the repair template smoke guard works.
• Flow table:
  • Reproduced: missing repair mode confirmed by source search/read.
  • Verified: pnpm install, rg audit, repair smoke plan generation, smoke incomplete-plan failure, final plan check.
• Browser check: N/A.
• Outcome: goal repair <expectation> now has a documented workflow and dedicated repair plan template.
• Caveat: no checker script enforcement added; by design, this starts with narrow source/template rules.
• Design:
  • Chosen boundary: generic repair command in goal; derived skills are repair targets, not duplicate command owners.
  • Why not quick patch: argument hint alone would not teach ownership, classification, proof, or safety.
  • Why not broader change: editing task/slate-plan now would duplicate the repair command before seeing a real repair invocation.
• Verified: source/generation/template smoke checks passed as described above.

Final handoff / sync:

• PR: N/A.
• Issue / tracker: N/A.
• Browser proof: N/A.
• Caveats: Biome ignores touched markdown/rule paths; no mechanical checker enforcement yet.

Timeline:

• 2026-05-25 Task goal plan created.
• 2026-05-25 Read pasted and local goal, task, slate-plan sources.
• 2026-05-25 Added goal repair <expectation> repair mode to goal rule.
• 2026-05-25 Added docs/plans/templates/goal-repair.md.
• 2026-05-25 Ran pnpm install for generated skill sync.
• 2026-05-25 Smoke-instantiated goal-repair and confirmed unfinished plan fails check-complete.mjs.

Reboot status:

Question	Answer
Where am I?	Closeout
Where am I going?	Final plan check, then final response
What is the goal?	Add goal repair <expectation> workflow support
What have I learned?	Goal had template creation but no expectation-repair mode; a dedicated repair template keeps self-improvement narrow.
What have I done?	Patched goal source, regenerated generated skill, added repair template, smoke-tested template/checker behavior.

Open risks:

• None for this workflow patch. Future real repair invocations may prove a checker/helper needs mechanical enforcement, but this change deliberately starts with source/template law.