writing-draft
Pass
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill configures lifecycle hooks that execute local Python scripts, including
phase-gate-guard.py,writing-outline-guard.py, andwriting-claim-id-guard.py, to manage workflow states and enforce document quality. - [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by ingesting external planning data.
- Ingestion points: The agent reads content from
.planning/PRECIS.md,.planning/OUTLINE.md, and section outline files in theoutlines/directory. - Boundary markers: No explicit delimiters or instructions are provided to the agent to treat the ingested outline content as data rather than instructions.
- Capability inventory: The skill has permissions to write files to the workspace, create tasks, and spawn autonomous agent teammates.
- Sanitization: No validation or escaping is applied to the content read from the planning files before it is processed by the model.
Audit Metadata