52 lines
2.6 KiB
Markdown
52 lines
2.6 KiB
Markdown
---
|
|
name: copilot-reuse-audit
|
|
description: "Use when auditing the last 30 days of persisted Copilot artifacts for reusable patterns worth promoting into the shared repo."
|
|
argument-hint: "days=<n> workspace=<optional path filter> sources=<optional csv>"
|
|
---
|
|
|
|
# Copilot Reuse Audit
|
|
|
|
Use this skill when you want a repeatable audit of local Copilot usage artifacts
|
|
to find patterns that should become shared resources.
|
|
|
|
## Procedure
|
|
|
|
1. Run `resources/scripts/audit-copilot-usage.sh` instead of manually hunting
|
|
through transcripts, memories, and publish logs.
|
|
2. By default, the runner excludes the `copilot-resources` repo root from
|
|
workspace-backed candidate discovery so the audit does not ask you to review
|
|
patterns that already live in the shared repository.
|
|
3. Review the generated `audit-summary.md` first to understand coverage and top
|
|
candidates.
|
|
4. Use `selection-manifest.tsv` and `pattern-details/*.md` as the approval
|
|
surface for each candidate.
|
|
5. When you want interactive triage inside Copilot, use the
|
|
`review-audit-candidates` prompt so each pending row is handled one at a
|
|
time and written back to `selection-manifest.tsv`.
|
|
6. Use the transcript prompt-cost proxy fields to prioritize repeated long
|
|
prompts that are likely worth turning into a shared resource.
|
|
7. Treat those prompt-cost fields as triage signals, not as exact billing data.
|
|
6. Before asking for a decision, explain the candidate's likely purpose,
|
|
concrete benefit, audit context, and score rationale. If evidence is thin,
|
|
call that out explicitly instead of asking the user to infer it.
|
|
8. Promote only candidates that map cleanly to a portable skill, instruction,
|
|
prompt adapter, agent, hook, script, or template.
|
|
9. After rows are approved, run `resources/scripts/prepare-audit-promotions.sh`
|
|
to create draft resources or staging notes in the audit directory.
|
|
10. Keep non-portable or repo-specific findings in audit notes rather than
|
|
forcing them into shared resources.
|
|
|
|
## Outputs
|
|
|
|
- `audit-summary.md` — high-level run summary and top candidates
|
|
- `candidates-report.tsv` — scored candidate list with source references
|
|
- `selection-manifest.tsv` — editable approval surface for promotion decisions
|
|
- `pattern-details/*.md` — per-candidate evidence bundles with benefit, context, score rationale, and caveats
|
|
- `draft-resources/` — generated draft resource files for approved portable rows
|
|
- `staging-notes/` — generated follow-up notes for approved rows that still need design work
|
|
|
|
## Notes
|
|
|
|
- The runner is macOS-first in this iteration.
|
|
- Audit history is stored per machine under `.local/audits/<machine-id>/` in the
|
|
repo checkout and is intentionally git-ignored. |