The Agent Skills Directory

[SAFE]: The skill implements strict boundary rules that explicitly prevent the coordinator from editing implementation code, performing reviews, or conducting debugging. This design minimizes the risk of unauthorized system modifications.
[COMMAND_EXECUTION]: The skill uses local CLI tools br (Bead Runner) and bv (Bead Viewer) to manage task states and inspect project graphs. These commands are integral to the skill's orchestration logic and are used within the local project scope.
[DATA_EXPOSURE]: Coordination state is maintained through local JSON files (.beads/STATE.json, .beads/HANDOFF.json). No exfiltration of sensitive system credentials or user data to external endpoints was detected.
[PROMPT_INJECTION]: The instructions include clear 'Hard Gates' and 'Red Flags' designed to enforce agent compliance with the orchestration protocol and prevent deviations from its intended purpose.

beo-swarm