The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/scaffold_docs.py uses subprocess.run to execute npx create-fumadocs-app@latest. While the product-id input is sanitized via slugification and the package manager is restricted to a specific list of choices, this still involves running shell commands based on user-provided parameters.
[REMOTE_CODE_EXECUTION]: The skill uses npx to dynamically download and execute the create-fumadocs-app utility from the npm registry. It also uses git clone to fetch remote source repositories for documentation generation. Both operations are central to the skill's functionality but rely on fetching content from external sources.
[INDIRECT_PROMPT_INJECTION]: The skill implements an "incremental authoring loop" where it reads content from a source folder (which could be a third-party repository) and uses it to generate documentation pages. This represents an attack surface where malicious instructions embedded in the source files could potentially influence the agent's behavior during the content generation process.
Ingestion points: Files in the source directory are read in scripts/inventory.py (metadata) and during the authoring loop in SKILL.md Step 6.
Boundary markers: The skill instructions do not explicitly mandate delimiters or "ignore instructions" warnings when processing these file contents.
Capability inventory: The skill has capabilities to write files to the filesystem (scaffold_docs.py, copy_assets.py) and perform network operations through delegated skills (lov-deploy-to-vercel).
Sanitization: There is no evidence of specific NLP-level sanitization for the content of the files being documented.

lov-repo2docs