Loaded when editing prompt files (agents, skills, CLAUDE.md, workflow markdown) or updating user-visible docs. Carries the editing judgment — what to delete, what to preserve, and where ripples land when a rename or restructure happens.
Bro's code-touching flow — verify world model, triage the requirement (reuse vs build), propose branch, write spec, dispatch SWE, verify on return, atomic-close. Loaded on the first code-touching ask of a session.
Bro's push-gate orchestration — reaping unsigned commits, spawning pr-reviewer per task, and the all-pass push + PR-create + post-merge cleanup path. Loaded by bro when the push hook blocks or the Human asks for review-before-push.
Bro's response when something fails — an MCP tool returns iserror=true (halt + surface), the trajectory-server is unreachable (degraded sqlite3 readonly fallback), or the world model is unavailable (retry, identify the lock holder, reconnect the binding). Loaded reactively on the first failure of a session.
PR-reviewer's diff-level push-gate review — check this task's diff against its spec, then write the validationrecord verdict that gates the push. Loaded by pr-reviewer when reviewing a committed task. Self-contained.
Generate a new project-local skill at .claude/skills/ /SKILL.md and attach it to existing agents. Loads when the user asks to capture a repeatable behavior — e.g. "create a skill that codifies ", "teach swe to also ", "make a skill for ", "we need a checklist when ". Extends the consuming agent's skills: frontmatter…
SWE's self-review heuristics — load only when about to atomic-close. Mechanical patterns (bare except, missing timeout, mutable defaults, etc.) are caught by scripts/hooks/code-quality-lint.sh; this skill carries the spec-fidelity and scope discipline judgment.