herbarium
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes project Markdown files to perform documentation consistency checks and reviews.
- Ingestion points: The skill ingests all files matching public globs in
.agents/herbarium.yml, includingREADME.md,AGENTS.md, and documents within thedocs/andskills/directories. - Boundary markers: There are no explicit delimiters or instructions for the agent to ignore potentially malicious prompts embedded in the documents it is reviewing or moving.
- Capability inventory: The agent is empowered to write, move, rename, and review documents, and executes the
de-novo skills herbarium checkCLI utility. - Sanitization: Document content is processed without identified sanitization, filtering, or instruction-stripping before agent review.
- [COMMAND_EXECUTION]: The skill executes the
de-novo skills herbarium checkcommand to perform linting and counting functions. This is a vendor-provided CLI tool associated with the skill's author ('de-novo') and does not involve remote code downloads.
Audit Metadata