The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local CLI tools including br, bv, and jq to manage project state and verify the structural integrity of planned tasks.
[DATA_EXPOSURE]: It accesses local project-specific files like CONTEXT.md and plan.md to confirm requirement coverage and plan soundness.
[PROMPT_INJECTION]: An indirect prompt injection surface is present as the skill ingests data from untrusted files (Ingestion points: CONTEXT.md, plan.md). There are no boundary markers or sanitization identified. The skill's capabilities include state modification via br commands. However, the mandatory human-in-the-loop approval requirement effectively mitigates the risk of these instructions being followed without oversight.

beo/validating