The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to execute the skills CLI tool from a public registry and fetches skill definitions from the vendor's repository at github.com/toy-crane/skills. These downloads are part of the skill's primary functionality for managing upstream dependencies.
[COMMAND_EXECUTION]: The instructions utilize shell commands to run the skills CLI and a project-local Python helper script (sync_companion_agents.py) for file synchronization.
[INDIRECT_PROMPT_INJECTION]: The skill processes agent definitions and metadata from a remote source and materializes them into the project's local agent directories. The implementation includes several security controls to mitigate risks associated with untrusted data ingestion:
Ingestion points: The sync_companion_agents.py script reads external manifests and agent source files from the .agents/skills directory.
Boundary markers: The script enforces a naming convention via regex and checks for duplicate identities during discovery.
Capability inventory: The script is restricted to local file operations (read, write, delete) and manifest updates.
Sanitization: The script validates that all file operations remain within the project root using the require_project_local_path function and employs an ownership lock file (.agents/toycrane-agents-lock.json) to identify and protect project-owned files from accidental overwrites.
[SAFE]: The Python synchronization script demonstrates defensive programming practices, including atomic file writes using temporary files and SHA-256 integrity checks to ensure the validity of managed agent files.

sync-toycrane-skills