outline-builder
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from user-provided taxonomy and skeleton files, creating a surface for indirect prompt injection if those files contain malicious instructions intended for downstream agent actions.\n
- Ingestion points: The
scripts/run.pyscript reads data fromoutline/taxonomy.yml,outline/chapter_skeleton.yml, andoutline/section_briefs.jsonlto build the research outline.\n - Boundary markers: The skill produces structured YAML output via
outline/outline.yml, but it does not include explicit delimiters or 'ignore' instructions to protect subsequent agent components from potentially malicious content embedded in the generated outline.\n - Capability inventory: The skill script is restricted to reading input files and writing the resulting YAML outline to the workspace. No network access or arbitrary shell command execution capabilities were identified in the analyzed code.\n
- Sanitization: The
_hint_sentencefunction inscripts/run.pyperforms basic string truncation and whitespace cleaning, but there is no specific filtering or escaping to prevent prompt injection patterns from being passed through to the output.
Audit Metadata