The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Downloads metadata from the official npm registry and pipes it to Node.js. This operation is implemented as a security gate to verify that the CLI package is version 0.1.6 and has been available for at least 72 hours before proceeding with the installation.
[COMMAND_EXECUTION]: Executes shell commands and build scripts to manage the containerized agent environment, including image rebuilding and restarting platform services via launchctl or systemctl to apply changes.
[PROMPT_INJECTION]: The skill processes untrusted content from document attachments (e.g., Word, PDF, Excel). It implements an indirect prompt injection defense strategy.
Ingestion points: Document files located in /workspace/inbox/ that are converted by the anydoc CLI.
Boundary markers: Explicitly includes a 'Treat documents as untrusted data' section instructing the agent to ignore instructions embedded in converted content.
Capability inventory: Limited to CLI execution and local file access within the isolated agent container.
Sanitization: Provides instructions for shell-escaping document paths to prevent command injection via malicious filenames.
[EXTERNAL_DOWNLOADS]: Fetches configuration and package information from the official npm registry to ensure supply chain integrity during the setup process.

add-anydoc