The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several Bash-based procedures to search the filesystem and manage files. It uses find, grep, sed, and git to identify broken references and orphaned files. These commands are typical for a codebase maintenance utility.
[EXTERNAL_DOWNLOADS]: The skill uses curl to verify the reachability of external URLs found in documentation. The implementation includes rate-limiting (sleep 0.5) to prevent abusive behavior and only captures HTTP status codes for validation purposes.
[PROMPT_INJECTION]: As the skill processes data found within the user's project files (specifically link targets and import paths), it possesses an indirect prompt injection surface. Maliciously crafted file content could theoretically attempt to influence the agent's behavior during the repair process, although the scripted nature of the procedures limits this risk.
Ingestion points: Reads markdown files and source code files (JS, Python, R) to extract links and imports.
Boundary markers: None explicitly defined in the provided scripts.
Capability inventory: Uses Bash, Edit, Write, and Read tools to modify the project structure and content.
Sanitization: Uses standard shell utilities like basename and realpath, but relies on the agent to correctly handle potentially malformed strings extracted from files.

repair-broken-references