build-verification-review
Skill1 starsChanged 2 months agoMITView on GitHub
Use after an Auto build phase ({step}b) claims deliverables are complete — before marking the phase DONE in {{docs.roadmap_path}}. Invoke: /build-verification-review. Mandatory in every overseer install: .cursor/rules/build-verification-required.mdc (alwaysApply). Agents must not mark DONE until this skill returns pass. This is the honesty gate: verify the implementation actually matches the frozen spec — not merely that tests are green or the agent said "done." Catch agent dishonesty: false completion claims, made-up behavior, tests that do not exercise real paths, docs/marketing that…
In this repository26 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .cursor/rules/build-verification-required.mdcCursor rule
- .cursor/rules/check-ok-thinking.mdcCursor rule
- .cursor/rules/governance-sync.mdcCursor rule
- .cursor/rules/l1-checkpoints-consumer-owned.mdcCursor rule
- .cursor/rules/no-docs-only-pr-to-main.mdcCursor rule
- .cursor/rules/orchestrator.mdcCursor rule
- .cursor/rules/print-next-closeout.mdcCursor rule
- .cursor/rules/tier-authority.mdcCursor rule
- .cursor/rules/workspace-authority.mdcCursor rule
- build-verification-reviewSkill
- check-okSkill
- deploy-verification-reviewSkill
- freeze-review-loopSkill
- freeze-reviewSkill
- governance-syncSkill
- print-nextSkill
- workspace-statusSkill
- check-okSkill
- deploy-verification-reviewSkill
- freeze-review-loopSkill
- freeze-reviewSkill
- governance-syncSkill
- knowtationSkill
- print-nextSkill
- workspace-statusSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 3 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 3
Use after an Auto build phase ({step}b) claims deliverables are complete — before marking the phase DONE in {{docs.roadmap_path}}.
- Distinctive: line 31
Before ROADMAP row → DONE
- Very specific: line 32
Before Tier-3 merge to {{vcs.git.main_branch}}
- Distinctive: line 46
Test files — do they assert real behavior or tautologies?
- Very specific: line 76
When honesty.enabled: true, a pass MUST be accompanied by appending (or confirming a prior append of) a verification_evidence entry with …
- Very specific: line 83
When honesty.enabled: false, ledger append is skipped.
- Very specific: line 97
Use the same escalation categories as freeze review: security, irreversible, real_money, gates_tier3.
- Very specific: line 106
Do not mark DONE in ROADMAP until pass
- Very specific: line 138
When {step}b completes, NEXT block should say:
- Very specific: line 147
No merge to {{vcs.git.main_branch}} on findings or blocked
- Distinctive: line 148
No waiving V8 ("tests pass therefore done") without addressing citations
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Build verification review skill
Use after an Auto build phase ({step}b) claims deliverables are complete — before marking
the phase DONE in {{docs.roadmap_path}}. Invoke: /build-verification-review.
Mandatory in every overseer install: .cursor/rules/build-verification-required.mdc (alwaysApply).
Agents must not mark DONE until this skill returns pass.
This is the honesty gate: verify the implementation actually matches the frozen spec — not merely that tests are green or the agent said "done."
Purpose
Catch agent dishonesty: false completion claims, made-up behavior, tests that do not exercise real paths, docs/marketing that overstate what shipped, and scope drift from the frozen contract.
This did not get lost — it was always the intent of SD-3 (Thinking → Auto). The kit shipped:
| Gate | When | Tool today |
| Freeze review | Before Auto build ({step}a) | freeze-review / freeze-review-loop + ok review --freeze |
| Build verification | After Auto build ({step}b) | This skill (thinking model; manual/opt-in) |
| Mechanical proof | During/after build | Seven-tier tests (policy/test-tiers.yaml) |
Tests prove code runs; this skill proves the code matches the frozen spec and the agent was honest about what landed.
When to use
- Auto phase
{step}bcomplete; tests reported green - Before ROADMAP row → DONE
- Before Tier-3 merge to
{{vcs.git.main_branch}} - Especially: video pipelines, billing, user-facing outputs — anywhere agents previously "approved" falsely
Model
Always thinking-high — independent reviewer; not the same session that did the build if
possible (fresh chat or explicit "verifier" role). Never Auto for this gate.
Inputs (read all)
- Frozen spec artifact (
frozen: truefrom{step}a) — ground truth {{docs.roadmap_path}}— phase deliverable row{{docs.handover_path}}— what the build session claims landed- Git diff vs feature-branch base (or
muse/gitlog for the build commits) - Test files — do they assert real behavior or tautologies?
ok status— footprint/drift if kit files touched
Verification checklist (every item needs evidence)
| # | Check | Dishonesty signal |
| V1 | Every frozen deliverable exists at the path the spec names | "Implemented" but file missing or stub |
| V2 | Public APIs match frozen interfaces (signatures, exit codes, fail-closed branches) | Spec says X; code does Y |
| V3 | Tests cover frozen test-matrix rows — not only happy path | assert True, empty tests, mock-only |
| V4 | No scope creep beyond frozen spec | Extra features not in contract |
| V5 | No silent deletion of frozen requirements | Spec requirement removed without spec update |
| V6 | Governance docs truthful | ROADMAP/HANDOVER say DONE but tests fail or deliverables missing |
| V7 | No secrets, injection surfaces, or unsafe defaults introduced | grep + read changed paths |
| V8 | Agent claims match verifiable state and (when honesty.enabled) are bound to ledger verification_evidence artifacts | "All green" with empty/unrelated diff; "tests passed" with no test_output hash; "deployed" / "healthy" with no deploy_health ref+hash; "UI verified" with no screenshot hash — or a claimed pass with no matching ledger entry when require_verification_evidence: require |
Evidence table (required in skill output whenever honesty module is enabled, and recommended always):
### Evidence
| type | sha256 | ref | notes |
| --- | --- | --- | --- |
| test_output | <64 hex> | <label or path> | <optional> |Rules for a skill verdict of pass (frozen process rules):
- V1–V7 unchanged in meaning.
- V8 requires citing verifiable git/test state as today.
- When
honesty.enabled: true, apassMUST be accompanied by appending (or confirming a prior append of) averification_evidenceentry withbv_verdict: pass, matchingphase_id+frozen_spec+round, and a non-emptyartifactslist that includes at least onetest_outputartifact whosesha256digests the test output the reviewer actually used. deploy_healthandscreenshotartifacts are required in the entry only when the build session's claims mention deploy/health or visual/UI proof; otherwise they are omitted. The skill must not invent fake deploy/screenshot evidence.- When
honesty.enabled: false, ledger append is skipped; V8 still requires claims↔git/test honesty in the review text (baseline without L2). findings/blockedrounds MAY appendverification_evidencewith the correspondingbv_verdictso the chain records failed rounds; this is allowed but not required for skill progress. A laterpassround is a separate append (newround).
Verdicts
| Verdict | Meaning | Next step |
pass | Implementation matches frozen spec; claims honest | Mark phase DONE; governance-sync; Tier-3 merge when ready |
findings | Cited gaps — fix on feature branch | Fixer addresses citations only; re-run this skill |
blocked | Escalating category or fundamental spec violation | Human required |
Use the same escalation categories as freeze review: security, irreversible, real_money,
gates_tier3.
Loop (bounded)
Same pattern as freeze-review-loop:
- Max rounds: 5 (build fixes should be smaller than spec fixes)
- Reviewer → findings → fixer (cited lines only) → re-review until
passor stop - Do not mark DONE in ROADMAP until
pass
Output format
## Build verification — <phase-id> round <N>
**Verdict:** pass | findings | blocked
**Frozen spec:** <path>
**Diff scope:** <commits or file list>
### Findings
| ID | Sev | path:line | Claim vs reality |
| --- | --- | --- | --- |
| BV1 | MAJOR | src/foo.py:42 | Spec §3.2 requires fail-closed; returns success on error |
### Honest summary (only if pass)
One paragraph: what actually shipped, backed by paths in the diff.CLI helpers (non-substitute)
./cli/ok status
# run the phase's seven-tier test command
./cli/ok governance-sync --dry-runCLI cannot replace thinking verification — run this skill after tests pass.
Handover integration
When {step}b completes, NEXT block should say:
Model: thinking-high.
Run build-verification-review against <frozen-spec-path> before marking <phase> DONE.Hard stops
- No merge to
{{vcs.git.main_branch}}onfindingsorblocked - No waiving V8 ("tests pass therefore done") without addressing citations
- Uncited findings invalid — re-review
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

