| name | audit-workspace |
| description | Check the workspace against its own standards. Find rules without enforcement, drifted ADRs, stale docs, and missing consequences. Run periodically. |
Audit Workspace
Usage
/audit-workspace
Overview
Lifecycle position: Utility/periodic — run periodically to check workspace
governance health. Not tied to the per-issue lifecycle.
Periodic governance health check — the "garbage collection" pattern. Verifies
that the workspace follows its own rules. Reports findings in the conversation.
Not the same as a structural validation script — those scripts check
structural config (repos match config files, projects are set up correctly).
This skill checks governance: are rules enforced? Are docs current? Are ADRs
still accurate?
Checklist
1. Principles enforcement
For each principle in docs/PRINCIPLES.md, check whether an enforcement
mechanism exists:
| Principle | Enforcement | Status |
|---|
| Human control and transparency | PR template consequence checklist | OK / Missing |
| Enforcement over documentation | Pre-commit hooks, CI checks | OK / Missing |
| ... | ... | ... |
Flag principles that exist only as documentation with no hook, CI check,
or guardrail.
2. ADR accuracy
For each ADR in docs/decisions/:
- Read the ADR's decision and consequences
- Verify the decision is still implemented as described
- Check that consequences listed have been addressed
- Flag any ADR whose status says "Accepted" but whose implementation has
drifted
3. Script reference table
Compare the script reference table in AGENTS.md against actual scripts
in .agent/scripts/:
- Scripts listed in the table that don't exist → stale reference
- Scripts that exist but aren't in the table → undocumented
- Descriptions that don't match the script's actual behavior → inaccurate
4. Template validity
For each template in .agent/templates/:
- Is it referenced somewhere (AGENTS.md, knowledge docs, skills)?
- Does it reference files that exist?
- Is it consistent with current conventions (e.g.,
.agents/ not
agent_context/)?
5. Consequences map currency
Read the consequences map in .agent/knowledge/principles_review_guide.md:
- Do the "If you change..." items still exist at the listed paths?
- Are there new high-impact files not covered by the map?
6. Instruction file consistency
Check that framework adapter files are consistent with AGENTS.md:
.github/copilot-instructions.md
.agent/instructions/gemini-cli.instructions.md
CLAUDE.md
Flag any rules in AGENTS.md that should be reflected in adapters but aren't.
7. Stale worktrees
.agent/scripts/worktree_list.sh
List any worktrees that appear abandoned (no recent commits, merged PRs).
Report Format
## Workspace Audit
**Date**: YYYY-MM-DD
### Summary
| Category | Findings |
|---|---|
| Principles enforcement | X of Y enforced |
| ADR accuracy | X of Y current |
| Script references | X stale, Y undocumented |
| Templates | X issues |
| Consequences map | X gaps |
| Instruction consistency | X issues |
| Stale worktrees | X found |
### Findings
#### <Category>
| Item | Status | Details |
|---|---|---|
| ... | OK / Issue | ... |
### Recommended Actions
- [ ] <specific action items>
Guidelines
- Report, don't fix — this skill identifies issues. Fixing them should
be separate issues with their own worktrees.
- Be specific — "AGENTS.md script table lists
generate_knowledge.sh
but it was removed in #274" is actionable. "Some references may be stale"
is not.
- Don't nitpick — focus on things that would confuse agents or humans.
Minor formatting inconsistencies aren't worth flagging.
- Run periodically — after a batch of PRs merge, or when starting a
new work cycle. Not after every commit.