Auto

discuss master orchestrator (v3)

Overview

4-stage cadence Stage ① master orchestrator delegating to 3 independent sub-workflows (bundled 3-tier independent clarification: strategic / phase / subtask):

sub	gate ref	mode	when fires
strategic	judgments.stage-routing.discuss-strategic-delegate.fires	parallel	new_project / new_milestone / new_feature / is_major_release
phase	judgments.stage-routing.discuss-phase-delegate.fires	parallel	open_decisions ≥ 2 / cross_phase_data_flow / scope_days > 1
subtask	judgments.stage-routing.discuss-subtask-delegate.fires	parallel	approaches ≥ 2 / core_algorithm / has_api_contract / error_cost=high

Engine runtime spawns 3 sub-workflow phases via runMasterOrchestrator per T3.5.W0.1 (sister sub-workflow runWorkflow SDK pattern, parallel fan-out via subagent default — Path A query() parallel OR Path B sub-shell fallback per T3.5.W2.1 dogfood cycle decision LOCK)。

Chain-isolation 铁律 (sister CLAUDE.md "Fallback 三条铁律")

• 拿不准 → 倾向跳过, 但在响应里透明声明: "这次跳过了 X, 因为 Y。如果你认为需要请明说"
• 用户明示 → 覆盖判据 (用户说 "先 brainstorm" / "跑 office-hours" / "讨论一下" 时无条件激活)
• 链式互不前置: 跳过战略层 ≠ 必须跳过 phase 层; 每层独立判断 (防止 "上层没跑下层不敢跑" 的死板)

Capability refs

Sister workflows/capabilities.yaml:

• planning-with-files — Bucket 4 核心 capability (impl: claude-code-plugin, cmd: /plan, discuss sink findings.md)
• sub strategic upstream → gstack-office-hours + gstack-plan-ceo-review
• sub phase upstream → gsd-discuss-phase
• sub subtask upstream → superpowers-brainstorming + grill-with-docs + grill-me

Invocation

• Slash command: /discuss <text> (bare per ADR 0030 namespace policy D-02 LOCK after harnessed setup)

How to invoke

Clarification needs real user dialogue, so run this stage directly in THIS session — do NOT spawn it, and do NOT improvise: follow these steps so the locked spec is persisted for the execution stages.

0. New-project bootstrap — if .planning/ROADMAP.md does not exist: invoke /gsd-new-project when that skill is available; otherwise create the minimal skeleton before continuing (then phase dirs follow .planning/phases/<NN>-<slug>/):
   • ROADMAP.md: # Roadmap + one table row per phase: | 01 | <slug> | <one-line goal> | in-progress |
   • STATE.md: # STATE (digest — keep <100 lines) + current: phase 01-<slug> / stage discuss + next: <action>
   • REQUIREMENTS.md: # Requirements + numbered rows - R1: <verifiable acceptance criterion>
1. Evaluate the clarification criteria for "$ARGUMENTS":
   • Strategic — new feature / milestone / unclear business scope → gstack /office-hours + /plan-ceo-review
   • Phase — ≥2 open implementation decisions / unclear cross-phase API contract → GSD /gsd-discuss-phase
   • Subtask — ≥2 distinct approaches / core algorithm / API contract design / high error cost → superpowers brainstorming
2. For each layer that fires, hold the dialogue with the user (use AskUserQuestion for option-style decisions) and lock every open decision. After locking blocking decisions, relay the deferrable set to the user in a single batched AskUserQuestion with each agent-recommended default pre-selected — only skip an item if the user explicitly defers it again; a deferrable item is never resolved without the user seeing it.
3. Transparent-skip layers that don't fire — state which were skipped and why.
4. Persist the locked decisions to .planning/phases/<NN>-<slug>/ via planning-with-files (findings.md / task_plan.md; NN = two-digit, one above the highest existing phase dir).

Output: a locked spec the execution stages (/plan → /task → /verify) consume without further user input.

<!-- harnessed-generated:v4.9.3 -->

References

• D-01 master orchestrator delegation pattern
• D-02 bare slash cmd convention (ADR 0030 namespace policy LOCK)
• D-04 Stage ① Discuss 三层独立判
• workflows/judgments/stage-routing.yaml — discuss-{strategic,phase,subtask}-delegate triggers
• workflows/discuss/{strategic,phase,subtask}/workflow.yaml — 3 sub-workflow Phase 3.4 SHIPPED
• workflows/judgments/fallback.yaml — 链式互不前置 chain-isolation 铁律