omh-prompt-import-readiness
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and audit untrusted prompt files from external sources (such as Claude Code or Codex libraries), creating a surface where malicious instructions embedded in those files could attempt to influence the agent during the import or normalization process.
- Ingestion points: External prompt directories (e.g.,
commands/), specific file paths provided by the user, and prompt files in various formats (Markdown, YAML/TOML frontmatter). - Boundary markers: The skill explicitly instructs the agent to use "not-evidence boundaries" and a "prepared-vs-observed boundary" to distinguish between the agent's internal guidance and the untrusted data being processed.
- Capability inventory: The workflow involves reading local file contents, listing directory structures, and checking internal command registries for naming collisions via the
hermesCLI. - Sanitization: The instructions mandate routing content to a dedicated safety review (
omh-security-safety-review) and generating a trust review artifact (prompt_trust_review/v1) to evaluate injection risks and secret leakage before any import occurs.
Audit Metadata