omh-codebase-onboarding
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from unfamiliar codebases.
- Ingestion points: Directs the agent to read various files within a repository (e.g., README, CI configurations, source code) to generate repo maps and glossaries.
- Boundary markers: The instructions mandate a clear distinction between observed evidence and inferred architecture/unknowns.
- Capability inventory: Uses search tools like
grep,ripgrep, andast-grep. The skill explicitly prohibits file mutation, dependency installation (e.g.,npm ci,uv sync), and running setup commands during the onboarding phase. - Sanitization: Includes explicit safety rules to redact secrets found in configuration or environment files and forbids exposing them in the output.
- [COMMAND_EXECUTION]: The skill references the use of an internal tool
omhfor recording runtime status. This is a standard operational command within the oh-my-hermes ecosystem and does not involve unsafe user input or external exfiltration.
Audit Metadata