Tooluck logoTooluck
SKILL.md

harness-audit

Agent skill fromundefined-ui/second-brain-os

Audit an agent's harness against the four rings: containment, guides, sensors, permissions. Use when the user asks whether their agent setup is safe, wants a review of sandboxing / permissions / hooks / AGENTS.md guardrails, or asks what a bad session could break. Read-only: reports the blast radius and a hardening list, changes nothing. Do NOT use for context layout (context-audit) or cost routing (gate-check).

How to install harness-audit

install harness-audit
# install the second-brain-os pack with the skills CLI
npx skills add undefined-ui/second-brain-os

# or add only this skill to Claude Code: copy the folder

plugins/agents-course/skills/harness-audit/

# into

~/.claude/skills/harness-audit/ (all projects)

.claude/skills/harness-audit/ (one project)

plugins/agents-course/skills/harness-audit/SKILL.md

MITFull file
1# Audit the four rings
2
3Theory: [The four rings](https://secondbrainos.dev/#/read/course-4-harness/the-four-rings)
4and [Harness practice](https://secondbrainos.dev/#/read/course-4-harness/harness-practice).
5Build from the outside in: containment, guides, sensors, permissions. The
6order matters because each ring must hold when every ring inside it fails —
7a guide can be ignored, a sensor can miss, an approval can be misclicked;
8a wall does not care.
9
10## Core rule
11
12Report and rank; never harden anything yourself. The audit's one organising
13question is the blast radius: describe the worst session possible under the
14current setup and what it costs. If the user asks you to apply fixes
15afterwards, that is a normal edit session, not this skill.
16
17## Workflow
18
191. **Ring one — containment: what can it physically reach?** Where does the
20 agent run: the user's checkout or a worktree/container? Which credentials
21 are in its environment — a read-only database user or the application's?
22 Open internet or an allowlist? Real API keys or test keys? The finding is
23 one sentence: "the worst session deletes X and spends Y."
242. **Ring two — guides: what does it read before acting?** Find CLAUDE.md /
25 AGENTS.md and tool descriptions. Every line should be a rule the agent
26 can act on — flag philosophy, history, and anything over roughly a page.
27 Flag missing rules for the repo's real landmines (migrations, generated
28 files, deploy scripts).
29…

Excerpt shown under the repository's MIT license.

More skills in second-brain-os

Keep exploring

Similar skill repositories

thedotmack /

claude-mem
116k

Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More

CCCXGHGM+135 skills

Newsletter

Join the Community

Subscribe to our newsletter for the latest news and updates