The Agent Skills Directory

[SAFE]: The skill implements a robust security and quality framework through its 'Checkpoints' system (CP0 to CP6.5). This includes mandatory task isolation via Git Worktrees for high-risk or autonomous tasks, which prevents experimental code from polluting the main repository.
[EXTERNAL_DOWNLOADS]: The skill performs version checks by fetching version information from the author's GitHub repositories (e.g., miles990/self-evolving-agent). It also fetches indices for other skills from dedicated repositories (claude-software-skills, claude-domain-skills). These operations are transparently documented and consistent with the functional requirements of a self-evolving system.
[COMMAND_EXECUTION]: The skill uses various shell commands for repository management (git), environment setup, and dependency installation (npm, pip, cargo). It also provides utility scripts for file indexing and skill creation. All command usage is presented as standard development practice and is not obfuscated.
[DYNAMIC_EXECUTION]: The 'skill-creator' module and the large codebase optimization scripts involve generating instruction files or shell scripts at runtime. These activities are clearly documented as part of the skill's purpose to automate development workflows and improve performance.
[PROMPT_INJECTION]: The skill provides instructions for the agent to analyze and execute user-provided goals. It mitigates potential prompt injection risks by using a 'North Star' anchoring system, mandatory checkpoints, and a multi-stage routing system that evaluates confidence levels before execution.

evolve