The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to spawn and manage multiple external CLI subprocesses (Claude Code, Codex, Gemini, Cursor Agent, OpenCode) using Node.js child_process modules.\n
This grants the skill broad access to the local system, file system, and environment variables of the host machine.\n
The 'Custom Engine' feature (engine: 'custom') allows for the configuration and execution of arbitrary binaries with user-defined arguments at runtime.\n- [PROMPT_INJECTION]: The skill contains potential surfaces for indirect prompt injection through its integration with external data sources.\n
Ingestion points: The --from-pr flag (documented in SKILL.md and tools.md) allows the agent to ingest content from GitHub Pull Request numbers or URLs into the active session context.\n
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions for content ingested from PRs.\n
Capability inventory: The skill possesses extensive capabilities including subprocess execution (claude, codex, etc.), file system modifications through those CLIs, and network operations via the embedded OpenAI-compatible bridge.\n
Sanitization: No explicit sanitization or validation of the PR content is described before it is interpolated into the agent's prompt.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of several external dependencies and tools from various public registries and sources.\n
Installation instructions in SKILL.md and getting-started.md reference packages from NPM (@enderfga/claw-orchestrator, @anthropic-ai/claude-code, etc.) and shell-piped installation scripts (https://cursor.com/install | bash).\n
The documentation explicitly instructs users to use the --dangerously-force-unsafe-install flag during installation to bypass security scanners that would otherwise flag or block the skill's subprocess spawning capabilities.

claw-orchestrator