Personal Claude Code skills.
Audit, diagnose, and improve existing Claude Code primitives — skills, subagents, prompts, .mcp.json, CLAUDE.md, or full .claude/ trees.
Triggers on: audit / review / diagnose / analyze / benchmark / test / "why isn't X working" + a named target.
Does not create new primitives from scratch — use prompt-creator or subagent-creator for that.
Workflow: identifies target → sets audit focus → gathers context in parallel → produces scored report (Trigger / Structure / Tokens / Coherence / Quality, each /5) → proposes fixes → dispatches rewrites to bundled builder agents (prompt-builder, subagent-builder) → re-measures → loops until done.
Run an intensive, facilitator-led brainstorming session for product/feature or technical/architecture problems — generating breadth of options, not analysis of a single answer.
Triggers on: brainstorm / ideate / generate ideas / explore options / "run an ideation session" + a named challenge, or /ultrabrain.
Does not use for stress-testing an existing plan (use grilling), reasoning through one hard problem (use ultrathink), reviewing code (use code-review), or authoring a skill (use generate-skill).
Workflow: frame the challenge in dialogue (How Might We + success criteria + constraints) → diverge widely with zero judgment across ≥3 techniques (inversion, analogies, SCAMPER, Six Thinking Hats) → converge (cluster, score, prioritize) → synthesize a prioritized ideas document plus a concrete action plan for the chosen idea.