The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its core functionality of fetching and processing external documentation from the web.
Ingestion points: scripts/crawl_docs.py and assets/templates/python/agentic_rag_docs_fetch.py fetch external HTML and Markdown content for agent processing.
Boundary markers: The fetching mechanisms do not wrap external content in clear delimiters or include system instructions to ignore embedded directives in the retrieved data.
Capability inventory: The skill includes file writing (scripts/crawl_docs.py), repository analysis (scripts/audit_repo_agents.py), and network retrieval tools.
Sanitization: The skill employs allowlists and prefix checks (e.g., ALLOWED_PREFIXES in agentic_rag_docs_fetch.py) to restrict network requests to official documentation domains.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to fetch documentation metadata and content.
scripts/crawl_docs.py and scripts/fetch_llms_txt_urls.py target langchain-ai.github.io and docs.langchain.com to seed its documentation cache.
These downloads target well-known, official service domains relevant to the skill's purpose.
[COMMAND_EXECUTION]: Several utility scripts execute system commands to perform analysis and snapshot library internals.
scripts/opensrc_snapshot.py uses subprocess.run to execute npx opensrc for downloading and inspecting library source code.
scripts/audit_repo_agents.py executes the rg (ripgrep) utility to scan the local repository for deprecated patterns.

langgraph-multiagent