The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external plan files (docs/plans/plan-{name}.md) which may contain untrusted data or instructions from previous agent steps. These files are used to trigger specialist agents and influence decision-making during the 'Auto-Fix Phase'.
Ingestion points: Reads docs/plans/plan-{name}.md, docs/adr/, and docs/plans/metrics/{name}.json.
Boundary markers: None explicitly defined to separate plan content from agent instructions during specialist reviews.
Capability inventory: The skill uses Task tools for background execution, Glob for file system checks, and writes to the file system (updating plans, creating ADRs, and writing metrics).
Sanitization: No explicit sanitization or validation of the plan file content is mentioned before it is passed to specialist agents (Opus/Sonnet).
[COMMAND_EXECUTION]: The skill utilizes the Task tool with run_in_background: true to spawn specialist reviewer agents. While these are part of the intended workflow, background tasks can sometimes obscure the agent's actions from immediate user oversight.
[SAFE]: The skill instructs the user to run /clear after validation but before execution, which is a good security practice to reset the agent's context and prevent instruction leakage between pipeline phases.

plan-pipeline-validate