pstack-maintain-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted project content which may contain instructions to subvert agent behavior.
- Ingestion points: Reads feature files in
.agents/skills/verify-*/and source code cited by the swarm tool. - Boundary markers: No boundary markers or 'ignore' instructions are present for processed content.
- Capability inventory: Can execute harness scripts, drive application state, edit verification files, and create pull requests.
- Sanitization: Lacks sanitization or structure validation for ingested files.
- [COMMAND_EXECUTION]: The skill executes project-local scripts as part of its live verification pass.
- Evidence: Step 4 (Live pass) instructs the agent to run harness scripts and exercise features using the target skill's launch model.
- [DYNAMIC_EXECUTION]: The skill can generate or modify executable harness scripts as part of its triage and correction process.
- Evidence: Step 5 (Triage) allows fixing harness gaps by making scripts executable and documenting their invocation, which are then re-driven live.
Audit Metadata