Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/jaqubowsky-audit-harness
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, security
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/audit-harness
What's inside
- Skill name
audit-harness- Description
- Review a coding session for friction and mistakes. Suggest the smallest harness changes, backed by evidence. Report only; nothing is edited.
- License
- MIT
- Instructions
- 28 lines in SKILL.md · view on GitHub
Files (4)
- SKILL.md
- extract.md · references/
- report.md · references/
- condense.jq · scripts/
SKILL.md (first lines)
# Audit harness
Improve the agent's environment at `{{root}}`, not its compliance score. End with proposals; changes are the user's call.
## 1. Read the session
Use the session the user names, or the current session by default. Session stores and extraction commands are in [references/extract.md](references/extract.md). Include child transcripts when the problem occurred in delegated work. Read enough surrounding messages to establish the task, outcome and what led to the problem.
## 2. Find friction
Look for mistakes, repeated commands, tool errors, user corrections, slow navigation, missing information and oversized tool results. Include instructions that were followed but produced a bad outcome, and the person's own habits that cost turns, such as a preference repeated in every steer or a setting on the Mac that stops the work. Each candidate needs a quote and source line, or a measured tool-call count. If the session shows no actionable problem, say so and stop.
## 3. Find the smallest fix
For each candidate, inspect the relevant harness source and check whether a solution already exists. Consult historical instructions only when the finding depends on what the session actually carried.
- Mechanical errors belong in deterministic checks. Inspect the project's declared checks, hooks and CI before proposing another one; an existing check that was not run may be the cause.
- Navigation and information gaps may need a pointer or access to an existing source.
- Instructions needing judgement may need deletion or clarification rather than another rule.
- Expensive tool calls may need narrower queries or output.
Prefer removing the cause over adding instructions or tooling. Check shared sources and both `pi/` and `claude/`; state when a proposal applies to only one. A candidate is ready when its evidence, cause, smallest fix and cost are known. Label an uncertain cause as unverified.
## 4. Report
Use [references/report.md](references/report.md). Order proposals by impact on safety and task completion, then wasted work. Merge candidates with the same cause. Show the report in the conversation. Save it only if the user asks, in the artifacts directory rather than the working tree. Stop without editing the harness.From jaqubowsky/fleet, MIT.
More from jaqubowsky/fleet
- 0.641score
diagnosing-bugs
jaqubowsky
Use when something is broken, throwing, failing or slow, when the user says "diagnose" or "debug this", or when the user describes a visual, layout or…
SkillAFreshness: A★ 13Claude Code +9 - 0.641score
check-feature
jaqubowsky
Walk the criteria of a finished change through the running app in a browser, one screenshot each.
SkillAFreshness: A★ 13Claude Code +9 - 0.641score
mockup
jaqubowsky
Use when the order asks to see how a change, a screen, a flow or a component could look, work or behave before it is built.
SkillAFreshness: A★ 13Claude Code +9 - 0.637score
orchestrating-agent-sessions
jaqubowsky
Fleet containers: put one up, steer it, read its task directory, land its branch, close it.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
brain
jaqubowsky
Answer a question from the personal knowledge base (LLM wiki) at ~/my-knowledge-base.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
analyze-task
jaqubowsky
Turning a ticket, bug report or feature request into an evidence-backed analysis before any code is written.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
babysit-pr
jaqubowsky
Answering an open pull request round after round.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
to-tickets
jaqubowsky
Break work down into tracker tickets.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
brain-dump
jaqubowsky
Capture a source into the second brain's inbox (raw/inbox/) for later ingestion.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
transcript
jaqubowsky
Use when a recording has to become text - a local audio or video file (mp3, m4a, wav, mp4, mov...), a YouTube or other online video link, a lecture, podcast…
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
implement
jaqubowsky
Building the slices of an analysis Plan into working code, one commit per slice.
SkillAFreshness: A★ 13Claude Code +9 - 0.635score
pr
jaqubowsky
Writing a pull request body.
SkillAFreshness: A★ 13Claude Code +9
Signals
- Stars
- 13
- Forks
- 1
- Open issues
- 0
- Last commit (this folder)
- Oct 21, 2026 (8 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/jaqubowsky/fleet
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from jaqubowsky
- 0.641score
mockup
jaqubowsky
Use when the order asks to see how a change, a screen, a flow or a component could look, work or behave before it is built.
SkillAFreshness: A★ 13Claude Code +9 - 0.641score
check-feature
jaqubowsky
Walk the criteria of a finished change through the running app in a browser, one screenshot each.
SkillAFreshness: A★ 13Claude Code +9 - 0.641score
diagnosing-bugs
jaqubowsky
Use when something is broken, throwing, failing or slow, when the user says "diagnose" or "debug this", or when the user describes a visual, layout or…
SkillAFreshness: A★ 13Claude Code +9 - 0.637score
orchestrating-agent-sessions
jaqubowsky
Fleet containers: put one up, steer it, read its task directory, land its branch, close it.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
analyze-task
jaqubowsky
Turning a ticket, bug report or feature request into an evidence-backed analysis before any code is written.
SkillAFreshness: A★ 13Claude Code +9 - 0.636score
babysit-pr
jaqubowsky
Answering an open pull request round after round.
SkillAFreshness: A★ 13Claude Code +9
Cite this page
audit-harness skill. Agent Mart. https://agentmart.dev/skills/jaqubowsky-audit-harness (last updated 2026-10-24)