ai-engineering-mastery

Find deepening opportunities in a codebase — refactors that turn shallow modules into deep ones. Use when the codebase feels hard to change, bugs cluster in certain modules, adding features requires touching many files, or proactively every few days as a health check.

Collaborative design exploration before implementation. Use when starting a new feature or project, when the approach is unclear, or when multiple design directions exist. Unlike /grill (which aligns on a given plan), /brainstorm explores and creates the plan. Enforces a HARD-GATE: no code until design is approved.

Systematic debugging loop for hard bugs and performance regressions. Use when something is broken and you don't know why, when behavior is inconsistent, or when error messages are unclear. Follows reproduce → minimise → hypothesise → instrument → fix → regression-test.

Deep alignment session before coding. Use when starting a non-trivial feature, when requirements are ambiguous, before refactoring, or when the user's request could be interpreted multiple ways. Surfaces assumptions, presents tradeoffs, and produces a verified implementation plan.

Request and handle code reviews with technical rigor. Use after completing a feature or task, before merging. Covers both requesting reviews and responding to feedback — verify before implementing, push back when wrong, no performative agreement.

Test-driven development with red-green-refactor loop. Use when building features, fixing bugs, refactoring code, or when the user mentions TDD, test-first, or red-green-refactor. Builds one vertical slice at a time.

Break any plan, spec, or PRD into independently-grabbable issues using vertical slices. Use after creating a PRD with /to-prd, after an architecture review with /architect, or any time a plan needs to be broken into actionable pieces.

Turn the current conversation context into a Product Requirements Document. Use after a /grill session when the plan is clear, when the user asks to formalize a feature discussion, or before breaking work into issues with /to-issues. Does NOT conduct an interview — synthesizes what's already discussed.

Use before claiming any task is complete, fixed, or passing. Enforces evidence-based completion — no claims without fresh verification. Applies to tests, builds, bug fixes, and requirement fulfillment.

Gain system-level context before making changes. Use when working in unfamiliar code, before changes that touch multiple modules, when debugging cross-cutting concerns, or when the agent seems to be making narrow assumptions.

Ultra-compressed communication mode. Cuts token usage approximately 75% by dropping filler while keeping full technical accuracy. Use when token usage is a concern, during long sessions, when the user asks for concise responses, or during automated/CI workflows. Stays active until user says "stop caveman" or "normal mode".

Compact the current conversation into a handoff document so another agent can continue the work. Use when ending a coding session, switching tasks, when the context window is getting full, or before handing off to a teammate or another AI agent. Saves to OS temp directory, not the workspace.

Scaffold per-repo configuration for all other skills. Run once per project before using any other skill. Configures issue tracker, triage labels, doc paths, test framework, and creates CONTEXT.md and initial ADR. Supports Claude Code, Cursor, Windsurf, Cline, GitHub Copilot, Codex, Aider, and any AI tool.