Adapt AI Behavior for Task Optimization

Defines six distinct behavioral operating modes that change how an AI assistant approaches a task, communicates, and prioritizes: BRAINSTORM (early planning - asks clarifying questions, offers at least three alternatives, no code yet, uses mermaid diagrams), IMPLEMENT (writing code - fast execution, complete production-ready code with error handling, no tutorial-style explanations or unnecessary comments, explicitly "quality over speed, read all references before coding"), DEBUG (systematic hypothesis-test-verify troubleshooting that explains root cause and prevention, not just the fix), REVIEW (severity-categorized feedback - Critical, High, Medium, Low - with improved code examples and acknowledgment of what's done well), TEACH (fundamentals-first explanation with analogies, diagrams, and a practical exercise), and SHIP (a pre-deployment checklist covering code quality, security, and performance). The AI auto-detects the right mode from trigger phrases in the request - for example "what if" or "ideas" trigger BRAINSTORM, "not working" or "error" trigger DEBUG, "deploy" or "release" trigger SHIP - or a user can switch explicitly with slash commands like /brainstorm, /implement, /debug, and /review. The skill also documents 2025-era multi-agent collaboration modes: an EXPLORE mode for a dedicated discovery/Explorer agent that produces a discovery-report.json, a cyclic PLAN-EXECUTE-CRITIC (PEC) pattern that decomposes a task into a task.md plan, executes it in IMPLEMENT mode, then reviews it in REVIEW mode, and a MENTAL MODEL SYNC behavior for saving and reloading context summaries between sessions.

Use this skill when defining or tuning how an AI assistant should adapt its communication style and priorities across a project's lifecycle, from early ideation through production release, or when building a multi-agent system that needs distinct explorer, planner, executor, and critic roles. It is a behavioral and prompting specification, not application code - don't use it expecting runnable software, and don't force a single mode onto a task that spans several: a request that starts as BRAINSTORM often needs to hand off to IMPLEMENT once a direction is picked.

Input is the user's natural-language request or an explicit mode command (/brainstorm, /implement, /debug, /review). Output format is mode-specific: BRAINSTORM emits pros/cons option lists, IMPLEMENT emits a code block with a one-to-two-sentence summary, DEBUG emits a symptom/root-cause/fix/prevention breakdown, REVIEW emits a severity-tiered list, TEACH emits a fundamentals-to-example walkthrough, SHIP emits a checklist, and EXPLORE emits a discovery-report.json.

References a companion clean-code skill as the mandatory standard for IMPLEMENT mode's output, and mermaid for diagram rendering in BRAINSTORM mode. The PLAN-EXECUTE-CRITIC pattern is designed for multi-agent orchestration where separate Planner, Executor, and Critic agents each take one mode.

Teams configuring an AI coding assistant's behavior across a project lifecycle, or engineers designing multi-agent systems that need explicit, named roles - explorer, planner, executor, critic - rather than one undifferentiated agent persona.