The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites and saves it as local markdown files. If an attacker controls the documentation site being scraped, they could embed malicious instructions designed to influence the AI agent when it subsequently reads those generated files.
Ingestion points: references/scrape_docs.py fetches content from user-provided URLs using crawl4ai.
Boundary markers: Absent. The script writes raw markdown content directly to files without adding headers or delimiters that warn other agents to ignore embedded instructions.
Capability inventory: The skill uses uv run to execute Python scripts and playwright to control browsers. It has file-write capabilities via pathlib.
Sanitization: The script performs basic filename sanitization (regex filtering of special characters) but does not sanitize the actual markdown content before writing it to disk.
[COMMAND_EXECUTION]: The skill instructions frequently recommend using uv run --with crawl4ai, which involves dynamic environment setup and execution of the local Python script references/scrape_docs.py. While the script itself is visible and follows its stated purpose, the pattern of executing scripts that process network data requires user oversight.

jb-docs-scraper