The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on subprocess management to control the local environment.
File cli_anything/drclaw/core/daemon.py: Uses subprocess.Popen to start a Node.js server process from paths determined at runtime.
File cli_anything/drclaw/core/openclaw_daemon.py: Spawns background watcher processes using subprocess.Popen to monitor workspace events.
File cli_anything/drclaw/core/openclaw_bridge.py: Executes the openclaw CLI via subprocess.run to deliver notifications.
File install.sh: Attempts to create global symlinks in /usr/local/bin using sudo if standard permissions are insufficient, representing a privilege escalation attempt.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting and acting upon untrusted project data and conversation histories.
Ingestion points: Untrusted data enters the context via cli_anything/drclaw/core/conversations.py (message retrieval) and chat.py (websocket event consumption).
Boundary markers: None identified; data is processed as shell arguments or JSON payloads without explicit delimiters to prevent command or instruction injection.
Capability inventory: Significant capabilities including starting/stopping servers and executing bridge commands are available in daemon.py, openclaw_bridge.py, and openclaw_daemon.py.
Sanitization: Relies primarily on standard library JSON serialization and Click's argument parsing, which may not prevent all forms of indirect instruction obedience.
[EXTERNAL_DOWNLOADS]: The skill requires several external libraries and assumes a local server infrastructure is already present.
File setup.py: Lists dependencies on click, requests, and websockets.
File install.sh: Executes pip3 install -e to perform an editable installation of the harness.

drclaw