Operator audit
Use this skill to inspect ACP audit events, telemetry, exports, replay evidence, and claim-safe operator traces.From its SKILL.md
npx -y skills add jasontang-ai/acp --skill operator-auditAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 2 commands, including `scripts/build-audit-summary.mjs` and 1 more.
SKILL.md
2.6 KB, 517 tokens by cl100k_base, as published. Nobody here has run it
What this skill is for
Use this skill to determine whether an ACP cycle has reviewable evidence for what happened.
When to use it
- Before declaring a demo, pilot, or benchmark run complete.
- When an export, replay, or lifecycle claim needs evidence.
- When a failure needs a preserved audit trail.
When not to use it
- Do not use it to optimize routing.
- Do not use it to infer missing metrics.
- Do not use it to mutate protocol history.
Inputs expected
- Cycle id or cycle JSON
- Desired export mode
- Claim boundary being evaluated
Preflight / prerequisites
- Confirm whether the review is source, generated evidence, or local debris.
- Confirm exports are generated through Relay, not hand-authored.
- Keep audit review separate from field-efficacy claims.
Workflow
- Fetch cycle, audit events, telemetry events, metrics, digests, routing decisions, and exports.
- Generate missing canonical exports if explicitly requested.
- Check for lifecycle anomalies, telemetry gaps, and missing output evidence.
- Produce a compact review summary with gaps and next actions.
Decision rules / judgment criteria
- Claims require artifacts, not memory.
- Missing telemetry is a gap, not a value to infer.
- Generated evidence must be classified before commit or freeze.
- Replay is an inspection path, not a semantic source of truth.
Escalation rules
Escalate when audit events contradict cycle state, export modes are missing, telemetry is absent for claim-bearing flows, or generated evidence is too broad to freeze safely.
Available scripts
scripts/build-audit-summary.mjs: inspect audit, telemetry, export, and lifecycle evidence from cycle JSON or a local demo cycle.
Outputs
- Audit timeline
- Telemetry summary
- Export readiness report
- Claim-safe evidence notes
Failure handling
Preserve failure outputs. Do not clean, rewrite, or regenerate evidence until the failure has been classified.
Trust / safety notes
Audit skills protect claim boundaries. They should make uncertainty more visible, not make the project look cleaner than it is.
Composition notes
Use inside pilot-cycle-review, foresight-submission-prep, and release evidence workflows.
Examples to inspect next
Read examples/casebook.md for audit success, missing export, ambiguous telemetry, and overclaiming examples.
Evaluation hooks
Run scripts/build-audit-summary.mjs and npm run skills:audit.
What ships with it: 5 files
3.4 KB alongside SKILL.md, 1 of them executable
checklists/
evals/
- rubric.md193 B
examples/
- casebook.md481 B
references/
- workflow-notes.md247 B
scripts/
- build-audit-summary.mjsruns2.3 KB