The Agent Skills Directory

Command Execution Surface: The skill instructs the agent to use a local command (npm run rai:sign -- -ProjectSlug {slug}) during Phase 6 to sign assessment artifacts.
Context: This command uses a variable ({slug}) derived from the session state. While this is a standard pattern for orchestrating local development utilities, it requires the agent to handle the variable safely to prevent command argument manipulation.
Indirect Prompt Injection Consideration: The skill is designed to capture and process potentially untrusted data from users regarding AI system definitions, stakeholder impacts, and risk indicators.
Ingestion Points: User input is collected during Phase 1 ('Capture and scoping') as documented in references/capture-coaching.md.
Capability Inventory: The skill performs file writing (e.g., rai-plan.md, artifact-manifest.json) and executes shell commands as noted in references/backlog-handoff.md.
Boundary Markers: There are no explicit instructions or delimiters provided in the reference files to warn the model about ignoring embedded instructions within the captured system descriptions.
Sanitization: The instructions do not specify a sanitization or validation step for the content captured from users before it is used in report generation or command arguments.

rai-planner