Don't make LLMs honest. Make every factual claim auditable. — An LLM Claim Auditing Layer with T1-T7 truth gradients. 98.1% business effectiveness on LiarBench v0.2.
SaferSkills independently audited anti-lie (Agent Skill) and scored it 96/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 1 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 1 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Installs the Anti-Lie verifier, OpenClaw hook, and shadow worker so outbound channel messages with concrete business claims are checked for evidence and stamped with PASS/WARN/BLOCK-style audit tails.
Activate this skill when the user or workspace mentions:
Do not use it for general fact checking without OpenClaw outbound-message audit needs.
From this skill directory:
bash install.shUseful options:
bash install.sh --dry-run
bash install.sh --port=17600
bash install.sh --workspace="$HOME/.openclaw/workspace"
bash install.sh --python=/path/to/python3The wrapper checks Python >=3.10, OpenClaw config, backs up current hooks to ~/.openclaw/.anti-lie-backup-<timestamp>/, dispatches to package/_install-linux.sh on Linux or package/_install-macos.sh on macOS, then runs verify.sh.
bash verify.sh
bash verify.sh --port=17600verify.sh checks package structure, Python/Node/OpenClaw prerequisites, service health for the host OS, /healthz, a neutral /verify call, and the bundled 13-case acceptance suite when the service is running.
Expected installed-state result: 13/13 acceptance PASS from the packaged verifier plus an overall success line.
bash uninstall.shThis stops Linux systemd user units or macOS launchd agents, kills the shadow worker/server processes, removes the installed anti-lie hook, and restores the newest ~/.openclaw/.anti-lie-backup-* hooks backup when present.
Environment variables used by the installer/runtime:
| Variable | Purpose | Default / Example |
|---|---|---|
ANTI_LIE_PORT | verifier service port | 17600 |
ANTI_LIE_VERIFY_URL | hook/worker verifier URL | http://127.0.0.1:17600/verify |
ANTI_LIE_SENTINEL_TARGET | target for audit-tail delivery | node/channel specific |
ANTI_LIE_TAIL_FALLBACK_MS | fallback wait before tail delivery | 30000 |
FEISHU_APP_ID | Feishu app ID for tail delivery | node specific |
FEISHU_APP_SECRET | Feishu app secret | node specific |
FEISHU_OPEN_ID | Feishu recipient open_id | node specific |
FEISHU_CHAT_ID | Feishu recipient chat_id | node specific |
See docs/INSTALL-GUIDE.md for the full node-by-node manual and troubleshooting commands.
Install Python 3.10+ first. macOS usually uses brew install [email protected]; Linux depends on the distribution package manager. Then rerun with --python=/path/to/python3 if needed.
Use another port consistently:
ANTI_LIE_PORT=17600 bash install.sh --port=17600
bash verify.sh --port=17600Linux: inspect systemctl --user status anti-lie-verify.service --no-pager and /tmp/verify_server-systemd.log. macOS: inspect launchctl print gui/$(id -u)/ai.openclaw.anti-lie-verify and ~/Library/Logs/anti-lie/verify-server.log.
Check the shadow worker and logs:
pgrep -af anti-lie-shadow-worker.js
tail -n 50 /tmp/anti-lie-audit.log
tail -n 50 /tmp/anti-lie-shadow.logThat indicates an old hook is still active. Reinstall this rev5 skill and confirm the active handler matches the package under hooks/anti-lie-verify/handler.js.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.