The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell command templates for archiving multi-file skills and managing file permissions using tar, gzip, and chmod. These are standard developer operations provided as instructional guidance for the user.
[REMOTE_CODE_EXECUTION]: It documents installation methods involving curl piped to tar for extracting skill archives. This procedure includes a mandatory integrity check using SHA-256 digests to ensure that downloaded payloads match the publisher's intended state before being processed by an agent.
[EXTERNAL_DOWNLOADS]: The instructions reference skill installers and discovery standards from trusted entities like Cloudflare and Vercel Labs, and include templates for downloading skill content from a publisher's controlled domain.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for agents to discover and install external skills (Ingestion: .well-known/agent-skills/index.json). It correctly identifies the attack surface of processing remote skill metadata and mitigates it through mandatory cryptographic verification (Sanitization: SHA-256 digests) and by discouraging reliance on unauthenticated third-party hosts. Capability inventory includes file-system writes to the agent's local skills directory and network-read operations via curl.

skills-visibility