source-to-skill

Pass

Audited by Gen Agent Trust Hub on Nov 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external documents (books, articles, transcripts) to generate behavior-shaping instructions for other agents. This creates a surface where malicious instructions embedded in source material could influence agent behavior.
  • Ingestion points: Local files, directories, and URLs are processed via the scripts/extract.py helper script and standard agent file tools, as described in the SKILL.md workflow.
  • Boundary markers: The skill includes a Source integrity gate that explicitly instructs the agent to treat source content as data rather than instructions and to ignore any embedded overrides or safety bypass attempts.
  • Capability inventory: The skill is authorized to write files to the .agents/skills/ directory and execute shell commands via its Python helper script.
  • Sanitization: The Path privacy gate and Provenance sections in SKILL.md mandate the redaction of local absolute paths and temporary directory information from any generated artifacts to prevent sensitive path exposure.
  • [COMMAND_EXECUTION]: The provided Python helper script (scripts/extract.py) uses the subprocess module to call system binaries for file conversion tasks.
  • Evidence: The script invokes pdftotext for PDF extraction and ebook-convert (from Calibre) for specialized e-book formats.
  • Safety Measures: The script implements internal safety limits for archive processing (e.g., MAX_ARCHIVE_MEMBERS, MAX_ARCHIVE_TOTAL_BYTES) to mitigate risks from malicious archives such as zip bombs.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 15, 2026, 04:37 AM