The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied URLs. 1. Ingestion points: scripts/fetch_page.py and scripts/render_page.py fetch HTML/JS. 2. Boundary markers: Data is structured for analysis but content is processed by the agent. 3. Capability inventory: The skill possesses file-write (SQLite and local files) and network-request (Google/Bing/Moz APIs) capabilities. 4. Sanitization: No content-level sanitization is applied to the analyzed text.
[DYNAMIC_EXECUTION]: scripts/render_page.py executes untrusted JavaScript from external sites using a headless Playwright instance to identify hydration shells and render content for audits.
[EXTERNAL_DOWNLOADS]: The toolkit downloads software and data at runtime from well-known services. scripts/unlighthouse_run.py uses npx to fetch and execute unlighthouse-cli from npm. scripts/sync_flow.py downloads prompt data and reference materials from GitHub (AgriciDaniel/flow).
[COMMAND_EXECUTION]: Various scripts use subprocess.run to call standard system utilities including git, whois, and exiftool. Sibling script execution is orchestrated via sys.executable.
[SAFE]: The skill uses high-quality security measures. scripts/url_safety.py centralizes SSRF and DNS-rebinding protection. scripts/google_auth.py manages sensitive OAuth tokens with restricted file permissions.

claude-seo