The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ingest.py executes a command using subprocess.run to integrate with the repo-search skill. This execution is performed safely by passing arguments as a list, which prevents shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill's primary function in scripts/clip.py involves fetching content from external URLs. It utilizes the requests library and trafilatura for clean text extraction. Additionally, it can interface with a local FlareSolverr instance to handle Cloudflare-protected pages.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection.
Ingestion points: Content is fetched from arbitrary, untrusted external URLs via scripts/clip.py and stored in ~/web-clips/.
Boundary markers: Extracted content is stored in markdown files with clear YAML frontmatter boundaries, providing some separation between metadata and external content.
Capability inventory: The skill has the ability to delete files (scripts/delete.py) and trigger external code execution via another skill's ingestion script (scripts/ingest.py).
Sanitization: Content is processed by trafilatura to extract article text and discard HTML boilerplate, which provides a layer of normalization, though semantic instructions within the resulting text are not filtered.

web-clipper