repository-wiki
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill includes explicit defensive instructions requiring the agent to treat all repository content, including ingested source documents and wiki pages, as untrusted data rather than instructions. This provides a strong baseline against both direct and indirect prompt injection attempts within processed documents.
- [DATA_EXFILTRATION]: The skill implements comprehensive path safety protocols. It forbids following symlinks, requires all paths to be validated within canonical roots, and explicitly prohibits the reading of secrets, credentials, or ignored files. All operations are local to the Git repository, and no unauthorized network or file access patterns were found.
- [COMMAND_EXECUTION]: The skill uses standard Git commands (such as
git hash-object) for integrity checks and file identification. These commands are used safely with non-dynamic arguments and are protected by a repository-wide mutation lock to prevent race conditions and TOCTOU (Time-of-Check Time-of-Use) vulnerabilities. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted source material but mitigates risks by requiring the agent to maintain strict boundaries between requests, evidence, and decisions. It uses metadata and manifest records to track source provenance and ensures that evidence records are labeled with their type and limitations, preventing the accidental elevation of untrusted content to confirmed facts.
Audit Metadata