cuozg/goal-verify

Goal Verify

Verify exactly one Docs/Goals/**.md goal. This is a gate, not implementation: re-check every acceptance criterion independently, audit goal-execute self-test evidence, then update the goal file with honest pass/fail evidence.

Inputs and contract

• Goal files come from goal-create: one file, frontmatter status completed, criteria under ## Acceptance criteria.
• goal-execute may add ## Verification evidence; treat that as evidence to audit, never as proof by itself.
• Verify one goal only. If no path is provided, find a likely unverified goal under Docs/Goals/;

Workflow

1. Read the full goal: frontmatter, desired behavior, key interfaces, criteria, plan, evidence, dependencies.
2. Set/keep status verifying while verifying; use blocked only for external blockers. Refresh updated when editing.
3. Build a private matrix: criterion text, expected proof, existing self-test evidence, independent check, result, severity.
4. Audit execution self-tests for each criterion:
   • relevant to exact criterion text
   • fresh or rerunnable now
   • reproducible command/tool/log/screenshot
   • strong enough to fail if behavior is broken
   • not skipped, stale, assumed, or hidden by compile/editor errors
5. Verify every criterion independently with the narrowest proof:
   • code: cite path:line
   • tests: run focused unit/Edit Mode/Play Mode checks
   • scene/assets: inspect GameObjects, components, serialized values, missing refs
   • runtime/editor: record deterministic observed state/logs
   • screenshot: only for visual criteria, preferably paired with state evidence
6. Decide each criterion:
   • PASS only if direct evidence proves the exact checkbox text.
   • FAIL/BLOCKED for UNCLEAR, stale logs, skipped tests, missing evidence, compile errors, assumptions, or invalid self-test.
7. Edit the same goal file in Verification part:
   • passed: - [x] ...
   • failed/unverified: - [ ] ... plus an indented callout directly below
   • update ## Verification evidence with per-criterion verdicts and accepted/rejected self-test evidence
   • set completed only if every criterion passes; otherwise pending or blocked
8. Report goal path, final status, criterion verdicts, self-test audit, commands/tools run, and next action.

Failure comments

Use required severity badges directly under failed verification evidence:

• > 🔴 = CRITICAL/HIGH: core criterion missing, test/compile fail, blocking runtime error, data loss/security risk.
• > 🟠 = MEDIUM: partial implementation, invalid/weak self-test, missing edge case, ambiguous/unproven behavior.
• > 🔵 = LOW/STYLE: non-blocking polish, docs/naming/style mismatch.
• > ✅ = PASS: criterion is met and evidence is provided.

Format:

[🔴 CRITICAL] [Criterion Text] (1 line summary) (Short explanations if needed)

• Evidence 1
• Evidence 2

Rules

• Preserve criterion text and goal-create/goal-execute sections.
• Do not implement fixes unless explicitly asked.
• Do not mark completed without independent passing evidence for every criterion.
• Do not trust self-test evidence until audited.
• Failed verifications must be commented in the goal file, not only in chat.