planning-setup
Pass
Audited by Gen Agent Trust Hub on Mar 28, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill implements a workflow to gather project-specific metadata (Project Name, Goal, Constraints, Key Questions) and interpolates these user-supplied values into templates. The resulting content is appended to the agent's persistent instruction files (
AGENTS.mdandCLAUDE.md), which define the agent's operational boundaries and logic. In the absence of sanitization or robust boundary markers, malicious input provided during the setup phase could compromise the agent's instruction set for the duration of the project. - Ingestion points: Gathered via user prompts during the 'Setup Intake' phase defined in
SKILL.md. - Boundary markers: None present. User inputs are placed directly into Markdown templates without delimiters or 'ignore' instructions.
- Capability inventory: Filesystem write/append access to project configuration and instruction files (
AGENTS.md,CLAUDE.md, and the planning directory). - Sanitization: No validation or filtering is applied to user-provided text before interpolation.
Audit Metadata