The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches configuration and instructions from Anthropic's official 'claude-plugins-official' and 'skills' repositories. These are verified trusted sources.
[COMMAND_EXECUTION]: Executes a local setup script (ensure_spec_repo.sh) to synchronize the Agent Skills Specification and utilizes claude install-skill to deploy the 'skill-creator' utility.
[PROMPT_INJECTION]: The research workflow (workflow-research.md) uses WebFetch to ingest content from the internet. This creates a surface for indirect prompt injection where malicious instructions on a webpage could influence the agent.
Ingestion points: references/workflow-research.md (WebFetch calls to arbitrary URLs found via search).
Boundary markers: Lacks explicit delimiters but includes strong instructions to prioritize official documentation and cite all sources.
Capability inventory: Writing to the filesystem (workflow-author.md), executing shell scripts (ensure_spec_repo.sh), and installing extensions (claude install-skill).
Sanitization: Mitigation depends on the multi-agent review process (workflow-review.md) and anti-hallucination constraints that require facts to trace back to cited documentation.

create-skill