MUST be activated when creating, updating, or evaluating a directory-specific AGENTS.md file through codebase reconnaissance. Use when the user asks to generate AGENTS.md documentation, audit an existing AGENTS.md for completeness, or stress-test an AGENTS.md against real tasks.
MUST be activated when deleting code (features, endpoints, jobs, modules, or functions) and doing cleanups, especially when removal can trigger cascading deletions or break externally reachable entry points.
MUST be activated when the user asks to write, edit, or review a Git commit message, including git commit, commit summaries, staged changes, or PR commit wording.
Applies default professional software engineering practices. Use for most coding tasks unless a more specific skill applies. Provides guardrails for minimal changes, safety, testing discipline, and uncertainty handling.
MUST be activated when the user asks for a reflection on the current conversation, including identifying which skills were used, suggesting general improvements to existing skills, and proposing new repeatable skills or commands.
MUST be activated when creating, editing, or maintaining agent skills (SKILL.md, skill structure, triggers, or workflows). Use whenever writing or revising a skill package, skill description, or activation behavior.