The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses gh repo clone to download the ginlix-ai/LangAlpha repository into the sandbox workspace for the purpose of proposing code fixes. This is a functional requirement for a self-improvement skill and targets the vendor's own official repository.
[COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) and git to create issues and pull requests. These operations are gated by a mandatory user consent requirement and a pre-submit checklist designed to prevent the accidental exposure of sensitive information.
[INDIRECT_PROMPT_INJECTION]: The skill operates on data encountered during other tasks (e.g., tool errors, malformed data from MCP servers). While this constitutes an attack surface, the skill implements significant mitigations, including mandatory human-in-the-loop approval, data sanitization instructions, and a focus on technical bug reporting rather than user-provided content.
[DYNAMIC_EXECUTION]: Proposing fixes involves running linting and testing tools (ruff, pytest) on local code. This is a standard development workflow and is confined to the sandbox environment.
[DATA_EXFILTRATION]: While the skill sends data to a remote repository, it contains explicit 'Safety Rules' and a 'Pre-Submit Checklist' that forbid the inclusion of tokens, secrets, API keys, or private user data in any issue or pull request.

self-improve