Harness Audit (Four Rings)
A read-only audit of your agent's harness across four rings — containment, guides, sensors, permissions — that names the blast radius of the worst possible session.
What it does
Instead of reviewing your code, this skill reviews the environment your agent runs in, as four layers of defence:
- Ring 1 — containment: what can the agent physically reach? Your checkout or a worktree/container, read-only DB user or the app's credentials, open internet or an allowlist, real API keys or test keys.
- Ring 2 — guides: are CLAUDE.md / AGENTS.md and tool descriptions made of rules the agent can act on? It flags philosophy, history, anything over roughly a page, and missing rules for the repo's real landmines (migrations, generated files, deploy scripts).
- Ring 3 — sensors: is there one command (
make checkor equivalent) running tests, linter and types, and does a hook run it automatically after edits? A check the model must remember to invoke is graded as a guide, not a sensor. - Ring 4 — permissions: what is auto-approved. Irreversible actions (push to main, deploy, spend, delete data, send messages) should require a human; reversible ones inside the walls should not. Approval fatigue counts as a hole, not a virtue.
Output is a one-sentence blast radius, held/open per ring, and a hardening list ranked outermost-first. The skill changes nothing.
Who it's for
- Developers running coding agents against real repositories who want to know how bad a bad session could get
- Teams whose repo holds production credentials, deploy scripts or migrations
- Platform/DevOps folks standardising hooks and auto-approve lists
Examples
- "Is my Claude Code setup safe?" → ring-by-ring verdict plus a prioritised hardening list.
- "What could one bad session break?" → a blast-radius sentence derived from your actual credentials, permissions and sandboxing.
- "Review my AGENTS.md and hook guardrails" → flags unactionable prose, missing landmine rules, and checks that never run automatically.
· · · Install guide · · ·
Try it now, no install
Paste this into Claude to use the skill without installing anything.
Read the instructions in this file and follow them to help me: https://raw.githubusercontent.com/undefined-ui/second-brain-os/HEAD/plugins/agents-course/skills/harness-audit/SKILL.md What I want: (describe your task here)
If Claude can't open the link, open it yourself and paste the contents instead.
↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.
Install in the Claude app (no terminal)
- Download the ZIP with the button below.
- In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
- Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
Install in Claude Code
Let Claude do it — paste this into Claude Code
Install the skill I found on Claude Skill Mart. Copy the plugins/agents-course/skills/harness-audit folder from the GitHub repo undefined-ui/second-brain-os into my ~/.claude/skills/harness-audit/. When it's done, tell me in one line what this skill can do.
Install with a command instead
git clone https://github.com/undefined-ui/second-brain-os.git /tmp/second-brain-os && mkdir -p ~/.claude/skills && cp -r /tmp/second-brain-os/plugins/agents-course/skills/harness-audit ~/.claude/skills/⚠ This is a third-party skill. Check the source repository before installing.
- Open a terminal.
- Clone the repository:
git clone https://github.com/undefined-ui/second-brain-os.git - Create the skills directory:
mkdir -p ~/.claude/skills - Copy the skill:
cp -r second-brain-os/plugins/agents-course/skills/harness-audit ~/.claude/skills/ - Verify that
~/.claude/skills/harness-audit/SKILL.mdexists. - Restart Claude Code and ask something like "audit my agent harness" or "review my sandboxing and permissions".
- Remember it is read-only: to apply the suggested hardening, start a separate editing session.