Use when creating or revising an implementation-ready plan from a governing specification, artifact, or approved request. Covers approach selection, coverage, verification, and conditional task breakdown; do not use for specification authoring, source changes, or reports.
Use when a task has underspecified requirements, unresolved design directions, durable concept naming, vague option comparisons, workflow/prompt/skill design, or dissatisfaction without a settled desired change. Investigate the task normally, preserve settled decisions, state reversible assumptions visibly, and surface costly open decisions as bounded choices. Do not use for simple factual questions, fixed-format assignments, debugging with an obvious next step, or direct execution/final-artifact requests whose relevant decisions are already settled.
Use when creating or revising a decision-ready specification from a request. Covers requirements discovery, settlement, acceptance criteria, and readiness; do not use for implementation planning, source changes, or reports.
Use when creating durable agent artifacts: specifications, implementation plans, research notes, implementation reports, review reports, bug reports, or failure reports. Trigger when an agentic workflow needs a persisted contract, reusable evidence, or handoff record. Do not use for inline review comments, commit messages, transient scratch notes, or ordinary status updates that do not need a durable Markdown artifact.
Use when a primary agent must coordinate an approval-gated specification, planning, or implementation workflow through delegated capabilities. Trigger for staged workflows that select spec-only, plan-only, spec-then-plan, plan-then-implement, or implement profiles. Do not use for direct specialist work that has no staged user dialogue or approval boundary.
Use when planning or implementing a large request with multiple responsibilities, building an entire software or configuration system from scratch, or rebuilding one step by step. Trigger when the user asks for a complete feature set or environment whose parts must be introduced and verified incrementally without losing the final scope. Do not use for a small local change that clearly follows existing structure, or by itself for ordinary architecture, refactoring, testing, prompt-design, or approval-workflow tasks.
Use when UI accessibility or usability must shape specification planning, implementation planning, implementation, code/design review, or behavior-preserving refactoring. Trigger on work involving WCAG, semantic HTML, ARIA, keyboard operation, focus management, screen-reader support, forms, errors, contrast, reflow, status messages, or UX heuristics. Do not use for purely visual taste changes, legal compliance advice, framework-specific accessibility manuals, generic skill-authoring, or non-UI work unless accessibility/UX behavior is part of the task contract.
Use when writing, revising, or reviewing prompts that will be consumed by another model, agent, subagent, reusable prompt template, or system that expands or passes prompts to a model. Trigger for model-facing command prompts, AGENTS.md files, agent instruction files, subagent handoffs, and reusable prompt templates. Use to define what the receiving model will actually see, place instructions in the correct layer, delegate existing skills, reduce prompt bloat, avoid unnecessary negative constraints, and specify clear output contracts. Out of scope: ordinary end-user answers, domain-specific coding guidance, general writing style advice, and security-specific prompt-injection analysis.