Security audit
Security checks for vulnerabilities and agentic risk
This skill is a straightforward delegation guide for sub-agents, with no hidden code or persistence, though users should be careful when delegating sensitive work.
Install only if you want your agent to delegate tasks to sub-agents. Avoid sending sensitive data unless needed, keep delegated context minimal, and be cautious with timeout recovery because killing a sub-agent can interrupt unfinished work and logged state may contain private details.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
## Verification
Never trust "done" without checking:
- **Code:** Run tests, check syntax
- **Files:** Verify they exist and have content
- **Data:** Spot-check 2-3 items
The markdown instructs the operator to kill a session and log the last known state, then retry the task, but it does not warn that terminating a running session may discard in-progress work or that logging state may capture sensitive information. Under the markdown-specific warning requirement, this is a safety-relevant behavior affecting user data and system integrity that should be disclosed.
No suspicious patterns detected.