deep-work
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The workflow involves the agent reading user annotations and external reference code to guide its implementation tasks. This creates a surface where instructions embedded in processed files could influence agent behavior, though it is the intended primary function of the skill.
- Ingestion points: User-provided annotations in .claude/plan.md and external reference code provided during the planning phase.
- Boundary markers: Absent; the agent is instructed to identify any manual additions to the plan document as overrides or constraints.
- Capability inventory: File system access within the workspace and execution of development tools like linters.
- Sanitization: None; the agent is directed to address every annotation as a valid directive.
- [COMMAND_EXECUTION]: The skill utilizes the shell to run project-specific type checkers and linters during the implementation phase. This is characterized as a common and legitimate use of the agent's capabilities in a software development context.
Audit Metadata