The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the cligentic component library hosted on GitHub (Railly/cligentic) and provides instructions for its installation via bunx. This is described as a modular, opt-in mechanism for scaffolding CLI infrastructure with explicit user consent requirements.
[INDIRECT_PROMPT_INJECTION]: The skill provides thorough documentation on the risks associated with processing data from third-party APIs. It establishes a mandatory security practice to treat external response text as untrusted and to escape it before presentation to an agent to prevent instruction injection.
[DATA_EXFILTRATION]: The instructions include robust data protection strategies, such as mandatory redaction of secrets (passwords, tokens) before they are written to audit logs and the use of environment variables or OS keychains for sensitive credentials rather than local flat-file storage.
[COMMAND_EXECUTION]: The skill facilitates the design of CLIs that perform network and filesystem operations. It mitigates risk through a 'trust ladder' architecture, which gates high-consequence operations behind human approval, signed intent tokens, and dry-run validations.
[SAFE]: The skill serves as a defensive architectural guide, incorporating security best practices derived from real-world case studies. It includes references to vendor-related research repositories (e.g., crafter-research/sunat-cli) that align with the author's established resource patterns.

cli-build