agentsclimarketplace

Fable audit

Skill deepklarity/harness-kit/.claude/skills/fable-audit

A kit for building with AI agents and also the engineering patterns around it.

Install
npx -y skills add deepklarity/harness-kit --skill fable-audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Audit the fable roadmap (docs/fable_roadmap/) — verify claimed progress with proof, write a short report, update RESUME state + BACKLOG rank. Use at wave close or whenever the user asks how roadmap development is going. Triggers on - 'fable audit', 'roadmap audit', 'how is the roadmap going', 'audit progress', or /fable-audit.

SKILL.md

2.3 KB, as published. Nobody here has run it

Fable roadmap auditor

Trust nothing; verify everything. Lean by user directive: an audit is one short report plus two file updates — not a documentation pass.

  1. Verify claims with proof. For each task DONE since the last audit, run its verify step or inspect the merged artifact. Delegate command runs and file checks to parallel haiku subagents; keep judgment in the main conversation. The board is truth (board_overview.py / spec_trace.py / task_inspect.py --brief from taskit/taskit-backend/); doc-vs-board divergence is a finding. Downgrade anything unprovable.
  2. Recompute what has data: run the suites; count autonomy/unsticks from board history (hand-counted until the metrics script lands — say so). Report gaps as gaps.
  3. Grade against SCORECARD.md — every score move names the anchor (0/5/10) its evidence supports; recompute the overall (plain average). Scores live only in the fable_roadmap.md table — a score anywhere else is a finding. Then judge the ladder against fable_roadmap.md exit criteria; check the ratchet rules there. Violations are findings.
  4. Write audits/YYYY-MM-DD*.md: verdict line (on track / drifting / blocked + why), verified-done table with proof, demotions, metrics snapshot, top-3 risks with mitigations, ladder call. Readable in 30 seconds from the top.
  5. Update the living docs: RESUME "Current state" + BACKLOG re-rank (the top IS the next wave draft) + the ladder Status column if a level changed.
  6. Artifact only if the user asks — redeploy to the URL in RESUME's durable facts; never mint a new one.
  7. --dispatch only: convert the backlog top into a wave spec under bootstrap/ and load it (pre-dispatch checklist in the loader templates).

End with: verdict line, ladder level, top 3 next actions — plain prose.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.