swcc

Ministry executor - implement a specific sub-task by writing actual code changes. Use to execute individual work items from the State Council task list.

Party Committee - synthesize consultation reports into a final authoritative implementation decision. Use after consultation phase to produce the definitive execution plan.

State Council - decompose implementation plans into parallelizable sub-tasks for ministry agents. Use before execution phase to create a task dependency graph.

Discipline Inspection Commission - multi-mode oversight agent. Modes: post-decision review, post-task review, specialist execution review (correctness/design/standards), and cross-challenge. The caller specifies which mode via prompt.

Right-wing CPPCC member - propose conservative stable solutions with minimal changes. Use during consultation phase to generate low-risk proposals that maximize code reuse.

Policy Research Office - pre-consultation brainstorming agent. Explores problem space, identifies ambiguities, proposes 2-3 approaches with trade-offs. Use after triage (zhongban) and before consultation (CPPCC debate).

Think tank - adaptive research agent that serves any stage of the pipeline. Automatically assumes the appropriate institutional identity (社科院/发改委/工程院/审计署) based on the caller's needs. Use whenever an agent needs external research: SOTA solutions, official docs, open source libraries, API references, security standards, etc.

Central General Office - triage incoming tasks and assess scale (小/中/大). Use as the first step in any orchestrated workflow to determine consultation depth.

Centrist CPPCC member - synthesize left and right proposals into a balanced practical plan. Use for large tasks after zuopai and youpai have produced their proposals.

Left-wing CPPCC member - research SOTA solutions and propose bold innovative approaches. Use during consultation phase to generate ambitious proposals with code diffs.

Inspection only: run Discipline Commission code review and automated tests on current working tree changes. Use after writing code to check quality before committing.

Emergency mode: skip all consultation, execute all agents in maximum parallelism, fast verification only. Use for urgent bug fixes or tasks where the solution is already clear.

Consultation only: run brainstorming + CPPCC left-right debate and Party Committee decision without code execution. Use when you want to see the plan before committing to execution.

Pre-task brainstorming: explore problem space, identify ambiguities, propose approaches, get user confirmation before formal consultation. Use when you want to think through a task before committing to a plan.

Think tank research: directly invoke the 智库 agent for on-demand research. Supports 4 modes — 社科院 (SOTA research), 发改委 (feasibility analysis), 工程院 (implementation reference), 审计署 (compliance standards). Use when you need to research a technology, compare solutions, check docs, or investigate best practices.

Full pipeline democratic centralism orchestration: task triage → brainstorming → CPPCC consultation → Party Committee decision → inspection → execution → multi-dimensional review. Use when you have a coding task that needs multi-agent planning and execution.

Execute only: run State Council task decomposition, Discipline Commission task review, Ministry execution, and multi-dimensional review. Use when you already have a plan (from /xieshang or your own) and want to execute it.