The Agent Skills Directory

[SAFE]: The skill operates as a thin driver for a local code review engine. All identified patterns follow standard development and evaluation practices.\n- [COMMAND_EXECUTION]: The skill executes standard development tools (git, gh, glab, bb) to resolve diffs and post comments. These are gated by user-provided arguments and environment availability, following the principle of least privilege.\n- [EXTERNAL_DOWNLOADS]: External network operations are limited to fetching metadata from well-known forge providers (GitHub, GitLab, Bitbucket) via their official CLI tools. These operations are intended for PR analysis and do not involve untrusted code execution.\n- [REMOTE_CODE_EXECUTION]: No remote code execution or dynamic code evaluation patterns were detected. The evals harness uses a deterministic, LLM-free matcher for scoring findings against ground truth.\n- [DATA_EXFILTRATION]: No evidence of data exfiltration was found. Findings are reported to the user or posted back to the originating forge PR as requested by the user.\n- [PROMPT_INJECTION]: Instructions focus on operational flow and do not contain bypass markers, override instructions, or jailbreak attempts. The skill explicitly delegates safety and behavior guidelines to a shared engine.

delve