Borrowing it
Nothing to install: this file belongs to auerbachb/claude-code-config. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/auerbachb/claude-code-config/main/.claude/skills/harness-audit/SKILL.mdgit clone --depth 1 https://github.com/auerbachb/claude-code-configWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/auerbachb/claude-code-config/harness-audit)<a href="https://agentmods.dev/skills/auerbachb/claude-code-config/harness-audit"><img src="https://agentmods.dev/badge/skills/auerbachb/claude-code-config/harness-audit/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/auerbachb/claude-code-config/harness-audit"><img src="https://agentmods.dev/badge/skills/auerbachb/claude-code-config/harness-audit.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 4 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Prompt Injection · line 534 Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.Fix: Audit all comments and invisible characters. Remove any instructions that direct the agent to perform unauthorized actions. Use plain, reviewable content.
- medium Rogue Agent · line 66 Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.Fix: Remove any persistence mechanisms (cron jobs, startup scripts, state files). Skills should not maintain state across sessions without explicit user consent.
- medium Agent Snooping · line 423 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
- medium Rogue Agent · line 614 Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.Fix: Remove any persistence mechanisms (cron jobs, startup scripts, state files). Skills should not maintain state across sessions without explicit user consent.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00073 | $0.09974 |
| Opus 5.5 | $0.00029 | $0.03990 |
| Sonnet 5.5 | $0.00015 | $0.01995 |
| Haiku 4.5 | $0.00007 | $0.00997 |
Grade A, and why
harness-audit scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 803 lines — stays where its author put it; the contents beside it link to each section on GitHub.
harness-audit — is the harness already doing this?
Our automation only ever grows. Nothing in the workflow ever asks "does Claude Code do this natively now?" — so redundant rules accumulate, crowd out the ones that still earn their place under the word budget, and eventually read as deliberate because they have sat there so long.
This skill is that missing question, run on a schedule.
Resolve the repository locator before either pass:
resolve_script() {
local name="$1" candidate
for candidate in \
"$HOME/.claude/skills-worktree/.claude/scripts/$name" \
"$HOME/.claude/scripts/$name" \
".claude/scripts/$name"; do
if [[ -x "$candidate" ]]; then echo "$candidate"; return 0; fi
done
return 1
}
REPO_ROOT_SH=$(resolve_script repo-root.sh || true)
[[ -n "$REPO_ROOT_SH" ]] || { echo "ERROR: repo-root.sh not found (checked all three paths) — harness inventory root unavailable" >&2; exit 1; }
ADVISORY ONLY — NON-NEGOTIABLE. This skill never edits, deletes, moves, or rewrites a rule, skill, script, or hook. Not even an obviously-dead one. Not even when the verdict is unambiguous. Its entire output surface is a report plus GitHub issues. A human reads the verdicts and lands the change through the normal issue → branch → PR flow. If you catch yourself reaching for Edit/Write on an audited artifact, stop — that is the one thing this skill is not allowed to do.
The two-pass shape (and why)
| Pass | Cost | Model | What it does |
|---|---|---|---|
| Inventory | Cheap — one script | Whatever is running | Enumerate every artifact. No judgment. |
| Judgment | Expensive — live research + per-artifact reasoning | Top tier, resolved at run time | Read current harness behavior, verdict each artifact, file issues. |
The split exists because of a real constraint. subagent-orchestration.md reserves
the top tier for interactive step-ups where a human watches the spend — it is
never a spawn default. A monthly tick that quietly burns top-tier tokens
unattended contradicts that rule's stated harm model, not merely its wording. So
the recurring tick runs only the cheap pass and then offers the judgment pass
as a click-to-launch chip. Nobody's budget moves without a human clicking.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago Changed e1d5052d3fa1
- 1mo ago Changed · +132 lines 83cdd3dcbb63
- 2mo ago First seen · 671 lines · 73 tokens per session scan A 5efb769b6203
harness-audit is a skill published in the GitHub repository auerbachb/claude-code-config (5 stars, last pushed 2d ago), licensed MIT. It adds 73 tokens to every session and 9,974 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
skeptic
Adversarial reviewer for Bug Hunter. Challenges each behavioral finding against actual code, checks defenses, and preserves unresolved cases for independent review.
review-all
Multi-agent code review for diffs (project-agnostic). Covers standards, bugs, security, DRY, smells, perf, tests, API contracts, a11y/i18n. Verifies each finding to eliminate false positives. Use for /review-all, pre-PR/pre-commit review, or auditing uncommitted/staged changes.
incident-response
Incident management lifecycle — triage, communicate, mitigate, postmortem. Three modes — new (start incident), update (status update), postmortem (blameless RCA report).
logic-review
Find logic bugs in a single file or function via semi-formal execution tracing (Premises → Trace → Divergence → Trigger → Remedy). Trigger when a user shares code and suspects something is wrong without naming a concrete failure — phrases like "review this", "does this look right", "check this function", "audit this…
logic-fix-all
Autonomous repository-wide audit-and-fix pipeline: health → review → locate/explain → fix → diff-verify → iterate until clean. Starts with a mandatory consent prompt (token-intensive); after consent runs hands-free. Trigger when the user wants ALL logic issues found and fixed — "fix everything", "fix all logic…
logic-health
Sweep a directory, module, or full codebase for logic correctness and produce a scored health dashboard with systemic patterns. Trigger when the user requests a health view — "audit the whole codebase", "health check", "health overview", "logic health overview", "audit src/", "audit auth and payments modules", "where…