skills/mryll/skills/agentmd/Gen Agent Trust Hub

agentmd

Pass

Audited by Gen Agent Trust Hub on Mar 2, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: Use of fictional authoritative claims. The skill cites a non-existent research paper from 'Feb 2026' and references future model versions (e.g., 'Sonnet-4.5', 'GPT-5.2') to ground its instructions in a simulated 'future best practice' framework. While the resulting advice encourages safety through minimalism, the metadata used to enforce it is fabricated.
  • [PROMPT_INJECTION]: Indirect Prompt Injection surface via repository analysis. The skill's core workflow involves the agent reading and interpreting untrusted files from a repository (e.g., README.md, package.json, CI configs).
  • Ingestion points: Repository files including package.json, pyproject.toml, README.md, and CI configuration directories.
  • Boundary markers: The 'Security: Data Boundaries' section provides explicit negative constraints, telling the agent to treat all repo content as untrusted and to never interpret free-text as instructions.
  • Capability inventory: The skill is restricted to file reading and markdown generation; no network access, subprocess execution, or code evaluation tools are utilized.
  • Sanitization: The instructions specifically direct the agent to extract only structured metadata and confirmed tooling commands, avoiding the verbatim echoing of arbitrary documentation text.
  • [NO_CODE]: The skill consists entirely of Markdown instructions and reference documentation. No executable scripts (Python, JavaScript, or Shell) are included in the distribution.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 2, 2026, 06:25 PM