front-matter-writer
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from workspace files which could contain untrusted content.
- Ingestion points: The scripts/run.py file reads content from GOAL.md, outline/outline.yml, and citations/ref.bib.
- Boundary markers: Untrusted input is interpolated into static document templates using Python's format_map method.
- Capability inventory: The scripts/run.py script performs file-write operations to the sections/ and output/ directories.
- Sanitization: The skill implements a specialized _lint_reader_facing function in scripts/run.py that performs automated checks against the generated text to block internal process narration, deictic phrases, and forbidden stems, which serves as a mitigation against adversarial instructions.
Audit Metadata