The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data extracted from arbitrary web pages, which acts as an entry point for untrusted content. This content could potentially contain instructions aimed at manipulating the agent's downstream tasks.
Ingestion points: Data is gathered from external websites via browser_evaluate tools as described in SKILL.md and references/extraction-patterns.md.
Boundary markers: The workflows provided do not include specific delimiters or "ignore embedded instructions" markers when handling the extracted data.
Capability inventory: The skill can write files to the local system (fs.writeFileSync) and execute local Python scripts (scripts/process_results.py).
Sanitization: While the scripts facilitate converting data to JSON or CSV, there is no mechanism shown for sanitizing extracted text to prevent the agent from following embedded prompts.
[EXTERNAL_DOWNLOADS]: The skill script scripts/validate_urls.py imports the requests library, which is an external Python dependency.
This contradicts the requirement stated in SKILL.md that scripts should use the "Standard library only". The presence of this unpinned external dependency could lead to execution failures or supply chain risks if the environment is not pre-configured.

playwright-web-scraper