The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated scanners identified the domain witscode.com (specifically witscode.com/open-knowledge-format) as blacklisted. This URL is referenced as a community resource within references/bundle-variations.md. Furthermore, the file references/bundle-variations.md itself was flagged with a malware reputation hit by security scanners.
[COMMAND_EXECUTION]: The evaluation script eval/run_eval.py utilizes subprocess.run() to execute the skill's Python generator script during testing. This involves spawning shell processes from within the Python environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided markdown and YAML data ('bundles'), creating an attack surface for indirect prompt injection if the source data is malicious.
Ingestion points: The script assets/okf_site.py recursively scans and reads the contents of all markdown files within a user-defined directory tree.
Boundary markers: The processing logic lacks explicit boundary markers or instructions to disregard embedded commands in the source markdown.
Capability inventory: The skill possesses the capability to read files, write new project files to the disk, delete directories (shutil.rmtree), and copy bundle assets (shutil.copyfile).
Sanitization: While html.escape() is applied to metadata values in the emitted site's panel, the primary content of the markdown bodies is passed to the downstream Astro/Starlight build system without additional sanitization.

okf-site-kit