audit-progress
Use when an iteration has just finished and you need to verify behavior evidence quality in three tiers — deep evidence for current stories, impacted behavior for touched scenarios, and sentinel corpus for high-value regression detection.
- License 许可
- Apache-2.0
- 所属仓库
- zak-keown/moe
- 最近更新
- 2026-09-05
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- audit-progress
- description
- Use when an iteration has just finished and you need to verify behavior evidence quality in three tiers — deep evidence for current stories, impacted behavior for touched scenarios, and sentinel corpus for high-value regression detection.
02 资源文件 1
- 参考文档auditor-subagent-prompt.md3.6 KB
03 SKILL.md 正文
Auditing Progress
Overview
Runs after every iteration as part of the planning cycle. Verifies behavior evidence quality in three tiers using parallel adversarial review (PAR) — two paired auditor subagents evaluate the same work in parallel with competitive framing.
The audit answers: "Does durable, reusable evidence exist at the correct seam for every externally observable behavior this iteration touched?"
When to Use
Invoked by iterate after every run-iteration call, before picking the next iteration.
Audit Process
1. Partition the audit into three tiers
Read the per-epic requirement files in docs/moe/iterations/requirements/, docs/moe/iterations/behavior-scenarios.md, and docs/moe/iterations/behavior-corpus.md:
- Tier 1 — Deep evidence: stories marked
done:ITER-<current>and scenarios added or updated in this iteration. Audit every AC and its proof obligation thoroughly. - Tier 2 — Impacted behavior: all existing scenarios whose owning stories had code changes in this iteration (even if those stories were completed in earlier iterations). Verify the scenarios still pass.
- Tier 3 — Sentinel corpus: all scenarios with run cadence
sentinelin the behavior corpus. Compare against the pre-iteration baseline fromrun-iterationstep 3.
2. Dispatch paired auditor subagents (PAR)
Following the PAR methodology in skills/_shared/parallel-adversarial-review.md, resolved relative to this loaded document:
- Build the auditor prompt using
auditor-subagent-prompt.md. Include ALL THREE tiers:- Tier 1: full story cards with proof obligations + new/changed scenario cards
- Tier 2: impacted scenario cards + their current test results
- Tier 3: sentinel scenario IDs + baseline results + current results
- Resolve skills/_shared/par-reviewer-wrapper.md relative to this loaded document and wrap in its competitive framing
- Dispatch TWO auditor subagents in parallel
- Wait for both to return
3. Aggregate findings
Following PAR aggregation rules:
- Same finding from both auditors → one finding, high confidence
- Finding from only one auditor → separate finding, still actionable
- Severity disagreement → take the more severe assessment, always fix it
4. Process results
- If gaps found (any AC fails, evidence is too weak, sentinel regression detected):
- For AC failures: append gap stories to
requirements/(statuspending) or flip existing stories back fromdonetopending - For weak evidence: create evidence-improvement stories (add scenario, strengthen seam)
- For sentinel regressions: create regression-fix stories with CRITICAL priority
- Revise
roadmap.mdto add a follow-up iteration for the gaps
- For AC failures: append gap stories to
- If clean (all tiers pass, evidence is adequate):
- The iteration is confirmed done
- Return clean signal to the orchestrator
5. Return control
Return the audit result (clean or gaps) to the orchestrator. The orchestrator decides whether to loop or terminate.
Quick Reference
| Tier | What it checks | Failure means |
|---|---|---|
| Deep evidence | Every AC + proof obligation for current iteration | Story not done, evidence too weak |
| Impacted behavior | Scenarios whose surfaces were touched | Stale or broken scenario |
| Sentinel corpus | High-value journey scenarios | Regression in previously-working behavior |
| Reads | Writes | Dispatches |
|---|---|---|
requirements/, behavior-scenarios.md, behavior-corpus.md, product code/tests |
requirements/ (gaps), roadmap.md (new iteration) if gaps, behavior-scenarios.md (stale flags) |
Two auditor subagents in parallel (PAR) |
References
- skills/_shared/parallel-adversarial-review.md — PAR methodology; resolve relative to this loaded document
- skills/_shared/par-reviewer-wrapper.md — competitive framing wrapper; resolve relative to this loaded document
- skills/_shared/behavior-evidence-formats.md — scenario and proof obligation formats; resolve relative to this loaded document
auditor-subagent-prompt.md— auditor-specific prompt template
04 同仓库更多 Skills
write-skill
Use when creating new skills, editing existing skills, or verifying skills work before deployment
write-plan
Use when you have a spec or requirements for a multi-step task, before touching code
write-clearly
Apply Strunk's timeless writing rules to ANY prose humans will read—documentation, commit messages, error messages, explanations, reports, or UI text. Makes your writing clearer, stronger, and more professional.
windows-vm
Create, manage, or connect to a headless Windows 11 VM running in Docker with SSH access. Use when the user wants to spin up, stop, restart, or SSH into a Windows VM.
verify-completion
Use when about to claim work is complete, fixed, or passing, before committing or creating MRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always
using-moe
Use when starting any conversation - establishes how to find and use skills, requiring skill invocation before ANY response including clarifying questions