maintain-verification-skill
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret untrusted project data (source code and feature maps) to generate documentation and live verification recipes, which could be exploited to influence agent behavior.
- Ingestion points: Step 2 ('Source wave') and Step 3 ('Reconcile') involve subagents and the coordinator reading feature files and application source code from the project directory.
- Boundary markers: The instructions do not specify any delimiters or 'ignore embedded instructions' warnings when processing these external files.
- Capability inventory: The skill possesses the capability to execute 'harness scripts' and drive application instances during the 'Live pass' (Step 4), and to create Pull Requests that modify local project files (Step 6).
- Sanitization: There is no mention of sanitization or validation of the source code content or the generated 'live-verification recipes' before they are executed or committed.
- [COMMAND_EXECUTION]: The skill explicitly includes a 'Live pass' phase that executes local harness scripts and drives application processes.
- Evidence: Step 4 ('Live pass') describes driving instances of the app, and Step 5 ('Triage') requires that 'harness scripts' are executable and documented in the skill body for invocation.
Audit Metadata