The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bash and tmux to orchestrate parallel agent sessions. It builds and executes shell command strings to launch workers with specific tool restrictions and model parameters.
[PROMPT_INJECTION]: The skill processes untrusted input from fleet.json (Ingestion points: fleet.json; Boundary markers: none; Capability inventory: bash, tmux, git, python3 in launch.sh and worker-spawn.sh; Sanitization: regex-based validation for identifiers). While it implements regex-based sanitization for identifiers to mitigate command injection, fields such as worker tasks are interpolated into local JSON status files without explicit escaping, creating a potential surface for indirect prompt injection or schema confusion.
[SAFE]: Complex logic for independence validation and log formatting is implemented via Python and AWK scripts embedded directly in the source code. These scripts are executed locally and do not involve remote code downloads or external network calls. The skill maintains a local registry in ~/.claude/fleet-registry.json to track session state.

worktree-fleet