maintain-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses terminal tools, the
readlinkutility, and project-specific harnesses to 'drive' the application during the live verification pass. It also utilizes a task delegation tool to spawn concurrent research workers. - [INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted data by reading project source code and feature files to identify 'doc drift' and generate verification recipes.
- Ingestion points: The skill reads feature map READMEs, sibling files, and project-wide source code to explain user-facing features (SKILL.md Step 1 and Step 2).
- Boundary markers: Research sub-tasks are scoped with a brief to 'do not edit files, commit, or push', but there are no explicit delimiters or 'ignore' instructions for the ingested source code itself.
- Capability inventory: The skill can execute terminal commands, modify files within the verification skill's directory, and initiate pull requests via
link_pull_request(SKILL.md Step 4 and Step 6). - Sanitization: No explicit sanitization, validation, or filtering of the ingested source code content is performed before the data is reconciled and used to drive the agent's actions.
Audit Metadata