harness-audit
Skill1 starsChanged 41 days agoMITView on GitHub
Audits this bigin-skills plugin against current official Claude Code docs (skills, hooks, sub-agents, plugins, memory) — findings report only, never auto-fixes. Triggers: 'audit the harness', 'recheck against best practices', 'is this still current'.
26 other projects use a version of this.
Adopted by
Most recently taken up by
- hoshiori-dev/harness.mdtheir own version, first seen Nov 16, 2026
- lumlaolao/spike-evaltheir own version, first seen Nov 16, 2026
- gutohuida/AgentWeavetheir own version, first seen Nov 15, 2026
- 5uck1ess/devkittheir own version, first seen Nov 15, 2026
- venhdev/skillstheir own version, first seen Nov 8, 2026
- DaeSeokSong/oh-my-ouroborostheir own version, first seen Nov 1, 2026
Used with
Projects with it also use harness-design, project-init and 5 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
VictoriakaeySkillharness-design
Use when designing or improving the action space, observation format, tool boundaries, and evaluation signals of an LLM agent. Covers the five layers of harness design: what tools the agent has, what it sees after acting, how each tool is scoped, how behavior is evaluated, and how to iterate when the agent misbehaves. Prevents the most common harness failures — agents that pick wrong tools, ignore critical information, or loop without progress.
Also in 3 projects that use this
u9401066Skillproject-init
Initialize new projects using this template. Triggers: init, new, 新專案, 初始化, create project, 建立專案, bootstrap, scaffold project.
Also in 3 projects that use this
ibahgatSkillautopilot
Run autonomous multi-stage execution loops until acceptance criteria pass or blockers are reached.
Also in 3 projects that use this
Microsoft AzureSkilltest
Also in 3 projects that use this
nicolasestremSkillbrainstorm
Structured idea generation and solution exploration with systematic analysis.
Also in 3 projects that use this
MicrosoftSkillhandoff
Compact the current conversation into a handoff document for another agent to pick up.
Also in 3 projects that use this
Other versions26 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 26
- AnastasiyaW/claude-code-config★ 214same name
- AnastasiyaW/codex-claude-code-config★ 214same name
- Kanevry/session-orchestrator★ 53same name
- fabianmartinelli-fm/harness-audit★ 7same name
- fmslutions/harness-audit★ 7same name
- fernandopetry/keelson★ 5same name
- 5uck1ess/devkit★ 4same name
- gutohuida/AgentWeave★ 4same name
- DaeSeokSong/oh-my-ouroboros★ 2same name
- lumlaolao/spike-eval★ 2same name
16 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- skill-benchSkill
- ask-biginSkill
- bigin-harness-setupSkill
- contract-syncSkill
- debug-workflowSkill
- discovery-workflowSkill
- epic-workflowSkill
- flutter-figma-handoffSkill
- go-scaffoldSkill
- knowledge-distillSkill
- model-routerSkill
- napkinSkill
- next-scaffoldSkill
- nodejs-scaffoldSkill
- nuxt-marketing-scaffoldSkill
- nuxt-scaffoldSkill
- nuxt-ui-figma-handoffSkill
- project-scaffoldSkill
- session-handoffSkill
- sprint-distillSkill
- task-workflowSkill
- write-testsSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 22 edited versions we read keep these.
2
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 31
If any fetch fails, note it in the report and proceed with what's available — don't block the whole audit on one page.
- Distinctive: line 98
Ask: "Act on any of these now, or just log the report?"
- Very specific: line 102
If the user asks to act on specific findings, treat each one as its own task-workflow-scale change — spec gate for anything non-trivial, no …
“Common” is measured: kept by at least 6 in 10 of the 22 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
harness-audit
Full re-audit of this plugin against current official Claude Code docs. Findings report only — never fixes anything in the same run. Propose-then-stop, same discipline task-workflow and sprint-distill already enforce.
When not to use
- Auditing a target repo that
bigin-harness-setupscaffolded. This audits the plugin itself, not its output. - Measuring whether a skill actually improves outcomes — that's
skill-bench.
Anti-re-litigation rule: if .claude/audit-log.md exists, read it first. Any finding already listed there as Closed in a prior run — don't re-report it unless this audit independently finds something new about it (a regression, a doc change that reopens it). If the file doesn't exist, this is the first run; proceed normally and create it in Phase 4.
Phase 0: Pull current docs
Fetch and read directly — don't rely on training data:
- https://code.claude.com/docs/en/skills.md
- https://code.claude.com/docs/en/best-practices.md
- https://code.claude.com/docs/en/hooks.md
- https://code.claude.com/docs/en/plugins-reference.md
- https://code.claude.com/docs/en/sub-agents.md
- https://code.claude.com/docs/en/memory.md
If any fetch fails, note it in the report and proceed with what's available — don't block the whole audit on one page.
Phase 1: Audit dimensions
For each dimension: state what the docs say, what this repo currently does, cite the specific file/line, and classify as match / drift / opportunity. Skip anything that's already correct — the report only needs what doesn't pass.
Skip the generated table regions in CLAUDE.md/README.md (between <!-- gen:* --> markers) — they're mechanically derived from skills/*/SKILL.md + agents/*.md + tools/docs-manifest.json via node tools/docs_sync.mjs and are correct by construction. Any doc-staleness check focuses on the remaining manual surfaces instead: prose, the README tree diagram, and marketplace.json.
Skill frontmatter (all skills under skills/*/SKILL.md)
description:stays within the always-loaded budget in.claude/rules/skill-authoring.md(≤350 chars, ~250 target): one clause of purpose, 3–4 representative triggers, English only, no "Do NOT use" prose (that belongs in a body## When not to use).- Any skill whose side effects or timing sensitivity argue for
disable-model-invocation: true. - Any skill that should scope activation with frontmatter
paths:(skill-level — distinct from thepaths:already used in generated.claude/rules/*.mdtemplates). allowed-toolsused anywhere it should be, to pre-approve safe repeated tool calls.effort:set consistently — flag any skill that inherits from session by default where a pinned level would give more predictable behavior.- Any skill whose main phase is non-interactive and context-heavy (large git log/diff scans, sprint-scale reads) that should move to
context: fork.
Hooks
bash-guard.mjs,spec-gate-guard.mjs,injection-scan-guard.mjs,injection-gate-guard.mjs(including its stage-3 canary check), andcanary-seed.mjsregexes/exit-code/decision-output contract still match the current hooks.md schema (event names, stdin format, exit codes,hookSpecificOutputfields likeadditionalContextandpermissionDecision).- Any hook event from the current docs (
SessionStart,PreCompact,Stop, etc.) that's missing here but would close a real gap — not "adopt everything," only what fixes something concrete. bigin-harness-setup's settings.json merge logic (Phase 5-3) still matches how Claude Code actually merges hooks across scopes per current docs.
Sub-agents
- Anything in this harness's workflow (review, security scan, QA) that current
sub-agents.mdguidance suggests fits a.claude/agents/*.mdsubagent better than a skill phase. Check whether a read-only security-reviewer subagent (discussed early in this harness's design, never built) still makes sense against current docs, or whether a skill step remains the better fit.
Context budget & skill size
- Any
SKILL.mdover ~400 lines (of the 500-line cap) — name what should move toreferences/. - Any
references/*.mdfile not reachable via a link from its ownSKILL.md— orphaned content.
Plugin/marketplace structure
plugin.json/marketplace.jsonstill match currentplugins-reference.mdschema.- Any new plugin-level field (
agents,mcpServers, plugin-levelhooks) worth adopting over the current per-profile generation approach — only if it removes duplication, not novelty for its own sake.
Eval coverage
task-workflow/evals/evals.jsonexists but the harness itself was unreliable last run (see audit-log if present). Shouldbigin-harness-setup,nuxt-scaffold, orsprint-distillget the same should-trigger/should-not-trigger treatment, given they also carry "MUST use when" descriptions that could regress silently on edits.
Permissions
templates/merge/claude-settings.jsonand the seven profilesettings.jsontemplates (nuxt, nuxt-marketing, next, go, nodejs, flutter, tauri): anypermissions.allowentry broader than needed, or any common safe command missing that would cut permission-prompt friction.
Everything else
- Anything in the fetched docs genuinely new since the last audit (new frontmatter field, new hook event, changed discovery/routing behavior) that this repo doesn't use and plausibly should.
Phase 2: Report — STOP HERE
Output, then wait. Do not write or edit anything in this phase.
## Findings
| Dimension | Finding | Severity (blocking / drift / opportunity) | File:line |
|---|---|---|---|
...
## If I could only do one thing
<one paragraph>
## Closed since last audit
<items from audit-log.md marked Closed that this run reconfirmed, or "first run — no prior log">Ask: "Act on any of these now, or just log the report?"
Phase 3: Act (only on request, only what's approved)
If the user asks to act on specific findings, treat each one as its own task-workflow-scale change — spec gate for anything non-trivial, no batch-applying every finding silently.
Phase 4: Update audit log
Append to .claude/audit-log.md (create if missing):
## {DATE}
Docs checked: <list, note any fetch failures>
New findings: <count by severity>
Closed this run: <list, or none>
Deferred (logged, not acted on): <list, or none>This is what Phase 0 of the next run reads to avoid re-litigating closed items.
References
.claude/rules/skill-authoring.md— the conventions this audit checks compliance against.claude/audit-log.md(this repo, created on first run) — prior findings, closed/deferred statusskills/bigin-harness-setup/references/hook-guard.md— current guard scripts, for the Hooks dimensionskills/sprint-distill/SKILL.md— sibling propose-then-stop gate pattern this skill mirrors
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

