Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
No changes detected in this version. - Version bumped to 0.6.3. - Fixed creds
What This Skill Does
Tinman scans AI agent sessions for failure modes including prompt injection, tool misuse, and context bleed across 168 detection patterns. It intercepts tool calls before execution and returns SAFE/REVIEW/BLOCKED verdicts. Attack sweeps run 288 synthetic probes across 12 categories.
Agent self-protection via /tinman check lets an agent police its own tool calls before execution, catching attacks that external sandboxing alone would miss.
When to use it
- Block an agent from reading SSH keys or .env files before execution
- Scan the last 48 hours of sessions for prompt injection attempts
- Run a pre-deploy sweep to find privilege escalation vulnerabilities
- Set up continuous hourly monitoring for a long-running autonomous agent
- Review tool call verdicts in real time during a bash or file operation
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Scan the last 48 hours of sessions for security issues
- 1Run /tinman scan --hours 48 to analyze recent sessions
- 2Check sessions for prompt injection, tool misuse, and context bleed
- 3Classify each finding by severity S0-S4
- 4Write findings to ~/.openclaw/workspace/tinman-findings.md
- 5Run /tinman report --full to display results with evidence and mitigations
Findings report listing detected issues by severity, with evidence snippets and mapped mitigations for each