The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard shell commands including find, grep, tree, and wc to map repository structures and identify patterns like god files or layering leaks. It also utilizes git switch and git diff for managing refactoring branches, all of which are appropriate for a codebase management tool.
[EXTERNAL_DOWNLOADS]: The workflow references the execution of development tools such as npx jscpd and npm audit. These rely on the official Node Package Manager (NPM) registry, which is a well-known service. These tools are used for legitimate static analysis tasks like duplication detection and security auditing.
[DATA_EXFILTRATION]: The skill includes patterns to search for hardcoded secrets and credentials (universal-checks.md) and analyzes environment variable usage. This is performed locally as part of a security audit to identify risks within the target codebase. No network operations were found that would exfiltrate this discovered data to external servers.
[PROMPT_INJECTION]: The skill is designed to ingest and process entire codebases, creating a surface for indirect prompt injection if a file being audited contains malicious instructions. The skill mitigates this through its architecture, which mandates five distinct human-in-the-loop (HITL) checkpoints and rigorous multi-agent review and QA loops before any code is modified or committed.
[REMOTE_CODE_EXECUTION]: The implementation stage (quality-loop.md) involves subagents writing and executing test code. This dynamic execution is scoped to the repository being improved and is subject to the human-gated quality loop, ensuring that code changes are verified and approved by the user.

codebase-improver