The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Node.js scripts (create-unit.mjs, dismantle-unit.mjs, send.mjs) to execute system commands including git, herdr, gh, and trash. These operations are used to manage worktrees, branches, PRs, and agent lifecycles within the Herdr platform.
[DATA_EXFILTRATION]: The usage-state.mjs script accesses local usage metadata and session logs from ~/.claude/ and ~/.codex/ directories. This data is used to calculate model pool availability and pacing (burn rate) to optimize delegation. The data is processed locally and intended for internal orchestration logic rather than external exfiltration.
[PROMPT_INJECTION]: The skill identifies GitHub issue bodies as untrusted input that could lead to indirect prompt injection. It mitigates this risk by requiring the agent to flag issues from non-maintainers for explicit user confirmation before delegation and by using structured kickoff templates with clear outcome boundaries.

herdr-orchestrate