The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides an installation script (scripts/install-claude-code.sh) and several hook scripts (minions-directive.sh, minions-guard.sh, minions-sessionstart-sweep.sh) that are executed by the Claude Code environment to manage project-specific state and enforce orchestrator restrictions.
[COMMAND_EXECUTION]: The minions-wait.mjs script dynamically locates and executes the codex-companion.mjs script from the local Claude plugin cache using execFileSync to monitor background jobs.
[PROMPT_INJECTION]: The skill installs a hook that automatically injects a coordinator directive (references/directive.md) into every user prompt. This directive instructs the agent to override its standard behavior, prohibiting direct edits and mandating delegation to background subagents. This is the core intended functionality of the skill.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as part of its orchestration workflow:
Ingestion points: The coordinator agent reads and synthesizes results returned by background minion jobs via the minions-wait.mjs script.
Boundary markers: No explicit delimiters or boundary markers are implemented in the scripts to distinguish untrusted subagent output from coordinator instructions.
Capability inventory: The coordinator agent maintains access to powerful tools including the bash tool (for git management) and the Agent tool (for spawning further subagents).
Sanitization: No sanitization or validation of the data returned by background minions is performed before it is processed by the agent's main context.

minions