merge-forward
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill describes a 'session-end rebalancing' feature in SKILL.md and references/instantiate.md that allows for automated updates to the agent's own instruction files. While the process includes safety checks like invariant validation and recursion guards, the ability for a system to rewrite its own logic creates a risk of state corruption or malicious modification.
- [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading and acting upon data from the codebase and external review prose. 1. Ingestion points: The agent reads git logs, diffs, and review comments as described in the Workflow section of SKILL.md and gates.md. 2. Boundary markers: There are no instructions for using delimiters to separate untrusted repository content from system instructions. 3. Capability inventory: The skill has the ability to execute git merge, push to main, and docker compose commands as specified in SKILL.md and gates.md. 4. Sanitization: No sanitization is performed on ingested data to prevent it from influencing the agent's execution flow.
Audit Metadata