The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to probe the local filesystem to determine the environment configuration. Specifically, it checks for the existence of directories in the user's home folder (e.g., ~/.claude/plugins/marketplaces/ecc/ and ~/.claude/agents/) and identifies project metadata files (e.g., package.json, pyproject.toml, go.mod) to automate formatting and language detection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided plan documents to generate automated task descriptions.
- Ingestion points: The skill reads content from a file path provided by the user via the <plan-doc-path> argument.
- Boundary markers: Instructions define a specific structure for output (e.g., [Plan: <path>#step-<id>] and 'Acceptance' criteria), but the skill does not include explicit delimiters or instructions for the agent to ignore potentially malicious directions embedded within the source plan document.
- Capability inventory: The skill is designed to be generative only; it emits Markdown commands for user copy-pasting and is explicitly prohibited from invoking the /orchestrate tool or other execution tools itself.
- Sanitization: The skill instructions include escaping double quotes within task descriptions to ensure the final emitted command string is valid.

plan-orchestrate