audit-explainer

Explains your own OpenClaw security audit in plain English. It runs openclaw security audit --json on your machine (never --fix or --deep), lists the risky settings the audit reports by severity, shows which to look at first, and points you to OpenClaw's own fix guidance, so you know where to start hardening. It changes nothing, adds no checks of its own and doesn't review the code inside skills itself, beyond what OpenClaw's audit reports. The skill itself sends the output nowhere, but your agent's AI model will read it. Made by Is My Agent Safe (ismyagentsafe.com), run by Your Evenings Ltd. Is My Agent Safe is independent and isn't affiliated with or endorsed by OpenClaw or its maintainers. Free and open source (MIT-0). Optional $29 report on our website; the skill works fully without it.

Made by Is My Agent Safe (ismyagentsafe.com), run by Your Evenings Ltd. Free and open source (MIT-0). The optional full report on our website costs $29; the skill works fully without it.

Written with AI assistance by the Is My Agent Safe team at Your Evenings Ltd.

Last updated: October 20, 2026.

This skill explains OpenClaw's own security audit output. It isn't a scanner. It doesn't add checks of its own, and it can't promise anything about your setup beyond what the audit reports.

When to use it

Use it when the user asks what their OpenClaw security audit says, which findings matter most, or what to fix first.

What this skill does, and doesn't do

• It runs one command, locally: openclaw security audit --json
• It never runs --fix, --deep or any other command.
• It doesn't change config, permissions, plugins or skills.
• It doesn't write files. Don't save the audit output anywhere.
• It makes no network calls and sends the audit output nowhere.

Steps

1. Tell the user you're about to run openclaw security audit --json on this machine, and that it only reads. Then run it.
2. If the command fails or the output isn't JSON, say so plainly and show the error. Don't guess at results. Don't say "no issues found".
3. Read the JSON. The parts that matter:
   • summary: counts of critical, warn and info findings.
   • findings: one entry per finding, each with checkId, severity (critical, warn or info), title and detail. Some findings also have a remediation with OpenClaw's suggested fix.
   • suppressedFindings: findings the user chose to accept in security.audit.suppressions. They aren't in summary or findings.
   • secretDiagnostics: secret references the audit couldn't resolve. The audit carried on without them.
4. Write a short summary for the user (see "How to explain it").
5. Offer the next steps (see "Next steps").

How to explain it

Start with one line of counts, for example: "OpenClaw's audit found 1 critical, 2 warn and 3 info findings."

Then list findings in this order: critical, then warn, then info. Within each level, use the order from OpenClaw's own triage guidance (below). For each finding, give:

• the checkId and title, as written
• what it means, in one or two plain sentences, using the detail
• OpenClaw's own fix, quoted from remediation when there is one. If there isn't, say so, and point to the checkId in OpenClaw's audit checks reference.

Keep it short. Five findings is plenty for a first pass. Offer the rest if the user wants them.

Never repeat tokens, passwords, API keys or other secrets from the output, even if they appear in a detail. Write "[hidden]" instead.

The checkId prefix tells you the area (from OpenClaw's docs):

• fs.*: file and folder permissions
• gateway.*: Gateway bind, auth, Tailscale, Control UI and trusted proxy settings
• hooks.*, browser.*, sandbox.*, tools.exec.*: settings for each of those areas
• plugins.*, skills.*: plugins and skills you've installed
• security.exposure.*: who can reach the agent, combined with what its tools can do
• security.trust_model.*: shared context between users or between agents

OpenClaw's triage order, from its "Running the security audit" docs:

1. Anything "open" with tools turned on. Lock down DMs and groups first (pairing or allowlists), then tighten tool policy and sandboxing.
2. Public network exposure (LAN bind, Funnel, missing auth). OpenClaw says to fix this immediately.
3. Remote browser control. Treat it like operator access.
4. Permissions. State, config, credentials and auth files shouldn't be readable by group or others.
5. Plugins. Load only what you've chosen to rely on.

Also tell the user, when it applies:

• If suppressedFindings isn't empty, or you see the security.audit.suppressions.active finding: "Some findings are suppressed by your config, so they aren't counted here."
• If secretDiagnostics isn't empty: "The audit couldn't resolve some secret references, so those checks may be incomplete."
• Always: "This was a plain audit. OpenClaw only runs live Gateway checks and plugin-owned security checks with --deep, so those weren't included."

If there are no critical or warn findings, say: "No issues found in the checks openclaw security audit ran." Don't say anything more about the agent or setup than that.

Next steps

• OpenClaw's docs for each checkId, with its severity and main fix: https://docs.openclaw.ai/gateway/security/audit-checks
• What the audit checks: https://docs.openclaw.ai/cli/security
• OpenClaw also has openclaw security audit --fix. This skill never runs it. It changes file permissions and open group policies, but it doesn't rotate tokens or change network exposure. If the user wants it, they should read those docs and run it themselves.
• Suggest re-running the audit after any change.

Optional: our website check

Only mention this after the summary above, and only once. The user doesn't need it to act on anything above.

If the user wants a second view, they can open https://ismyagentsafe.com/check?utm_source=clawhub&utm_medium=directory&utm_campaign=evergreen&utm_content=audit-explainer-skill themselves. They run openclaw security audit --deep and openclaw --version in their own terminal, then paste both outputs on that page. Don't open the link, fill it in or send anything to it for them.

We check the text you paste. Your browser hides many common key and token formats before sending, but not all of them, so remove any other secrets first.

The check is free. It gives a red, amber or green result, with the most important finding first. Green means "no issues found in the checks we ran". If the check can't finish, the result is never green.

A full report is $29. It explains every finding in plain English, lists the fixes in order and includes a prompt for your coding agent. One-off $29, not a subscription. Refunded automatically if the check can't finish, or in full if you ask within 14 days.

Limits of the check: https://ismyagentsafe.com/legal/terms#limits-of-the-check

Support

Report bugs in this skill on its GitHub repo: https://github.com/ismyagentsafe/audit-explainer