ln-21-documentation-auditor

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process documentation and code comments which may contain untrusted data.
  • Ingestion points: The skill explicitly targets documentation entrypoints, public docs, maintainer docs, operational runbooks, and code-comment surfaces within the repository as specified in Step 1 of the checklist.
  • Boundary markers: The instructions do not provide explicit delimiters or "ignore embedded instructions" warnings for the agent when processing external documentation files.
  • Capability inventory: The skill utilizes file listing, text search, code search, Git operations (log, blame, diff), and shell execution (shell in non-mutating or dry-run mode).
  • Sanitization: There are no instructions for sanitizing or escaping the content of the documentation before the agent interprets or acts upon it (e.g., when verifying examples).
  • [COMMAND_EXECUTION]: The skill instructions permit the use of the shell to verify content found in documentation.
  • Evidence: The 'Tool Routing' table and Section 4 of the checklist allow for Shell in non-mutating or dry-run mode to verify help text, command existence, config parsing, and example syntax. While limited to non-mutating actions, this provides a mechanism where untrusted strings from documentation could be passed to a shell environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 05:02 AM
Security Audit — agent-trust-hub — ln-21-documentation-auditor