The Agent Skills Directory

[SAFE]: The skill provides instructions for a "dispatch loop" workflow designed to manage multi-step development plans. It emphasizes coordination rather than direct implementation.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool for standard software development tasks such as file system checks (ls), content searching (grep), version control analysis (git log, git diff), and test execution (pnpm test). These operations are consistent with the skill's stated purpose of verifying sub-agent work.
[EXTERNAL_DEPENDENCIES]: References are made to a plan_runner.py script. The documentation transparently notes that this script resides in the repository checkout of a related skill (/plan-run) and is not packaged directly with this skill's snapshot.
[INDIRECT_PROMPT_INJECTION]: The skill includes explicit security and reliability precautions against "hallucinated" or inaccurate sub-agent reports. It mandates that the primary agent perform independent verification (spot checks) and forbids sub-agents from further delegating tasks to prevent "orphan" tasks and lost context.
[DATA_EXPOSURE_AND_EXFILTRATION]: While the skill handles session transcripts and references image data (base64 restoration for screenshots), these actions are described as part of legitimate data recovery and task auditing processes within the agent's local environment.

dispatch-loop