deliver-pull-request
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from pull requests, including diffs, comments, and project configuration, which are untrusted external inputs (SKILL.md, Procedure 1).
- Ingestion points: Pull request head/base revisions, complete diffs, and consumer rules defined in external repositories.
- Boundary markers: The skill instructions explicitly require agents to 'Inspect untrusted hooks before execution; isolate them from secrets' (SKILL.md) and use structured checklists for readiness.
- Capability inventory: The skill has the authority to coordinate repairs (code edits), push changes, interact with CI systems, and merge pull requests (SKILL.md, Procedure 5).
- Sanitization: Instructions specify redacting reports for real exposures and sanitizing sensitive evidence before reporting findings (runtime-performance/REFERENCE.md).
- [DYNAMIC_EXECUTION]: The skill identifies a set of preferred packages from the '@vllnt' scope and provides instructions to 'query the registry and install the current release' at execution time (references/vstack/capabilities/project-alignment/references/default-packages.md).
- This behavior allows for runtime assembly and execution of code, though it is limited to specific vendor-owned namespaces.
- Mitigation: The instructions include a 7-day age gate for non-vendor packages and require verification of package status and README before use.
Audit Metadata