The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs read operations on local documentation and project files to verify internal references. It checks external URLs found in documentation using HEAD requests, which is a standard function for link verification. A security guardrail is implemented to exclude secret-bearing files (e.g., .env, .pem, .key) and gitignored files from being processed or exposed in the output.
[COMMAND_EXECUTION]: The tool utilizes local shell commands such as git diff, git apply, and which to interact with the repository and verify tool availability. All file system modifications via git apply are gated by a mandatory interactive user confirmation step.
[EXTERNAL_DOWNLOADS]: The skill references lychee, a well-known open-source link checking tool. It provides documentation on how to install it from official sources and uses it as an optional enhancement for URL verification, avoiding automated execution of remote scripts.
[PROMPT_INJECTION]: The skill processes documentation content which may originate from external contributors, presenting a surface for indirect prompt injection. Mitigation is provided through the support of 'ignore' blocks in documentation and the use of structured JSON output to clearly separate data from instructions for the host LLM.

oma-docs