The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon instructions from external plan files, which presents a vulnerability to indirect prompt injection if those files contain malicious instructions.
Ingestion points: The skill reads implementation plans from user-specified file paths such as Plan.md or files in docs/plans/ as described in the Workflow and Error Handling sections of SKILL.md.
Boundary markers: The skill instructions include a cognitive boundary by requiring the agent to 'review critically' and 'raise concerns' with a human partner in Step 1 before beginning execution.
Capability inventory: The skill allows the agent to modify files and run verification commands across the repository as dictated by the plan's steps.
Sanitization: There are no explicit automated sanitization or filtering steps mentioned for the plan content before it is interpreted by the agent.
[COMMAND_EXECUTION]: The skill instructions involve executing tasks and verification steps which typically involve shell commands.
Evidence: SKILL.md contains instructions to 'Run verifications as specified' and 'Follow each step exactly' (Step 2), which enables the execution of commands defined within the external implementation plan.

executing-plans