Normalize a consequential or ambiguous request when missing context, boundaries, or proof could materially change the work. Explicit invocation previews the improved prompt.
Explore requirements and approaches through collaborative dialogue, then write a right-sized requirements document. Use when the user says "let's brainstorm", "what should we build", or "help me think through X", presents a vague or ambitious feature request, or seems unsure about scope or direction -- even without explicitly asking to brainstorm.
Find root causes and fix bugs. Use for errors, failing tests, issue repros, stack traces, "debug this", or "why is this failing".
Create structured plans for multi-step tasks once the goal is clear enough to plan. Use after ce-brainstorm or ce-grill, or directly for clear planning requests. If the request has branchy product/scope ambiguity, run ce-grill first. After markdown plans, document-review runs before handoff.
Review recent code changes for bugs, regressions, product fit, conventions, performance, security, and blast radius.
Execute implementation work with a compact product-contract loop. Use when the user asks to build, fix, implement, polish, or finish a scoped task. Favor reasoning, prior art, smallest correct changes, focused tests, and real-surface proof. Do not run autonomous PR, CI, ticket, or residual-work pipelines.
Route project knowledge searches to relevant collections on this machine and classify evidence authority. Use for non-trivial framing and planning or when implementation intent, ownership, conventions, expected behavior, or history are unclear; skip exact grounded work. Delegate query and retrieval mechanics to the qmd skill.
Refresh stale .ai/solutions learnings against current code. Use for "refresh my learnings", audit stale learnings, or consolidate overlapping docs.