om-auto-merge-buddy
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from pull request descriptions, comments, and code diffs. To mitigate the risk of indirect prompt injection, it contains explicit defensive instructions for the agent to treat such content as data rather than instructions, and to report any detected directives (e.g., 'ignore previous instructions').
- Ingestion points: Pull request bodies, conversation comments, diffs, and CI logs enter the agent's context as untrusted data (referenced in
references/agentic-setup.md). - Boundary markers: The instructions explicitly define boundaries, directing the agent to disregard the 'untrusted content boundary' and quote suspected injections as evidence.
- Capability inventory: The skill possesses significant capabilities, including performing squash-merges, closing PRs, and executing local CI commands (
SKILL.md,references/automerge-now.md). - Sanitization: There are mandates to redact secrets and to quote rather than execute potential injection patterns.
- [COMMAND_EXECUTION]: The skill executes shell commands for local CI validation and project building. It limits these to 'operator-vouched' commands defined in the repository's configuration file.
- Evidence:
references/automerge-now.mdandreferences/agentic-setup.mdspecifically forbid the execution of commands extracted from untrusted PR content, restricting execution to local configuration defined by the project owner.
Audit Metadata