writing-plans
Pass
Audited by Gen Agent Trust Hub on Apr 11, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted requirement specifications to generate plans that include shell commands and source code.
- Ingestion points: Reads requirements and specifications from user-provided specification documents.
- Boundary markers: Employs a specific Markdown template and task structure to contain and delimit generated content.
- Capability inventory: Generates implementation tasks involving file creation, file modification, and shell command execution (e.g., git, pytest).
- Sanitization: Includes a multi-step review process where a specialized reviewer subagent validates the plan's contents and alignment with the specification before execution.
Audit Metadata