cruise-control（巡航控制）

Cruise Control Orchestrator

Run all StackShift gears sequentially without stopping between gears.

Activation

Activate when the user requests automatic, unattended, or hands-free StackShift execution. Common trigger phrases: "cruise control", "run automatically", "autopilot", "full workflow", "all gears".

Setup: Collect Configuration

Present the following prompts in order. Skip prompts that do not apply to the selected path (see skip rules below).

Prompt 1 - Route Selection: Ask the user to choose:

• A) Greenfield - Shift to new tech stack
• B) Brownfield - Manage existing code

Prompt 2 - Implementation Framework: Ask the user to choose:

• A) GitHub Spec Kit - Feature specs in.specify/, /speckit.* commands
• B) BMAD Auto-Pilot - Auto-generate BMAD artifacts from reverse-eng docs
• C) BMAD Method - Same docs, hands off to BMAD's collaborative agents
• D) Architecture Only - Generate architecture.md with your constraints
• E) Portable Extraction - Tech-agnostic business logic extraction
• F) Widget Migration - Legacy widget to React Router 7 + Iris

Prompt 3 - Clarifications Handling (Spec Kit only):

• A) Defer - Mark them, implement around them, clarify later
• B) Prompt - Stop and ask questions interactively
• C) Skip - Only implement fully-specified features

Prompt 4 - Implementation Scope (Spec Kit only):

• A) P0 only - Critical features
• B) P0 + P1 - Critical and high-value
• C) All - Everything
• D) None - Stop after specs are ready

Prompt 5 - BMAD Synthesize Mode (BMAD Auto-Pilot only):

• A) YOLO - Fully automatic, no questions
• B) Guided - Auto-fill + targeted questions

Prompt 6 - Architecture Constraints (Architecture Only only): Collect: tech stack, cloud provider, scale, hard constraints.

Prompt 7 - Portable Extraction Mode (Portable Extraction only):

• A) YOLO - Fully automatic
• B) Guided - Auto-fill + targeted questions

Prompt 8 - Widget Migration Mode (Widget Migration only):

• A) YOLO - Fully automatic
• B) Guided - Auto-fill + targeted questions Target stack is always React Router 7 + Iris + TypeScript (not configurable).

Skip rules:

• BMAD Method: skip prompts 3-6
• BMAD Auto-Pilot: skip prompts 3-4, 6-7
• Architecture Only: skip prompts 3-5, 7-8
• Portable Extraction: skip prompts 3-6, 8
• Widget Migration: skip prompts 3-7

After collecting all applicable answers, write them to .stackshift-state.json:

{
  "auto_mode": true,
  "auto_config": {
    "route": "<greenfield|brownfield>",
    "implementation_framework": "<speckit|bmad-autopilot|bmad|architect-only|portable-extract|widget-migrate>",
    "clarifications_strategy": "<defer|prompt|skip>",
    "implementation_scope": "<p0|p0_p1|all|none>",
    "synthesize_mode": "<yolo|guided>",
    "architecture_constraints": { ... }
  }
}

Log: "Configuration saved. Starting cruise control with [framework] on [route] path."

Gear Transition Protocol

After each gear completes, do the following before proceeding:

1. Verify expected output files exist using Glob or Read.
2. Update .stackshift-state.json: set currentStep to the completed gear name and append it to completedSteps.
3. Log: "Gear [N] ([name]) complete. Output: [list files]. Shifting to Gear [N+1]."
4. If any expected output is missing or a gear produced an error, stop execution immediately. Log: "Gear [N] ([name]) FAILED. Missing: [files]. Error: [message]." Ask the user: "Gear [N] failed. Retry this gear, or skip it and continue?"

Execution: GitHub Spec Kit Path

Gear 1: Analyze

Run the analyze skill. Pass the route and framework selections.

Execute AST analysis: node scripts/run-ast-analysis.mjs analyze.

Verify output: analysis-report.md exists and .stackshift-analysis/ directory was created.

Set implementation_framework to speckit in .stackshift-state.json.

If Gear 1 fails: check that the codebase directory is valid and contains source files. Ask user whether to retry or abort.

Proceed to Gear 2.

Gear 2: Reverse Engineer

Launch the stackshift:stackshift-code-analyzer:AGENT agent via Task tool. Extract documentation based on route (greenfield = business logic only, brownfield = business + technical).

Verify output: docs/reverse-engineering/ contains all 9 expected files including integration-points.md.

If Gear 2 fails: check docs/reverse-engineering/ for partial output. Report which files are missing. Ask user whether to retry or skip.

Proceed to Gear 3.

Gear 3: Create Specifications

Run the create-specs skill in automated mode. Generate the constitution using the appropriate route template. Create all feature specs programmatically. Create implementation plans for incomplete features. Set up /speckit.* slash commands.

Verify output: .specify/ directory exists with constitution.md and at least one spec under .specify/specs/.

If Gear 3 fails: check whether .specify/ was partially created. Report status. Ask user whether to retry or skip.

Proceed to Gear 4.

Gear 4: Gap Analysis

Run the gap-analysis skill (equivalent to /speckit.analyze). Identify PARTIAL and MISSING features. Create prioritized roadmap. Mark [NEEDS CLARIFICATION] items.

Verify output: gap analysis report exists in .specify/.

If Gear 4 fails: report the error. Ask user whether to retry or skip.

Proceed to Gear 5.

Gear 5: Complete Specification

Branch on the clarifications_strategy from configuration:

• If defer: skip this gear, log "Deferring clarifications." Proceed to Gear 6.
• If prompt: run the complete-spec skill interactively. Ask the user clarification questions. Resume after answers.
• If skip: mark unclear features as P2, log "Unclear features marked P2." Proceed to Gear 6.

If Gear 5 fails: report the error. Ask user whether to retry or skip.

Proceed to Gear 6.

Gear 6: Implement

Branch on the implementation_scope from configuration:

• If none: log "Specs ready. Stopping before implementation." Mark workflow complete.
• If p0: run the implement skill for P0 features only.
• If p0_p1: run the implement skill for P0 and P1 features.
• If all: run the implement skill for all features.

Use /speckit.tasks and /speckit.implement for each feature.

If Gear 6 fails: report which feature failed. Ask user whether to retry that feature, skip it, or abort.

Verify output: implemented features match the requested scope.

Log: "Cruise control complete. All gears finished for GitHub Spec Kit path."

Execution: BMAD Auto-Pilot Path

Gear 1: Analyze

Run the analyze skill. Set implementation_framework to bmad-autopilot in .stackshift-state.json.

Verify output: analysis-report.md exists.

If Gear 1 fails: ask user whether to retry or abort.

Proceed to Gear 2.

Gear 2: Reverse Engineer

Launch the stackshift:stackshift-code-analyzer:AGENT agent via Task tool. Extract all 11 documentation files including business-context.md and decision-rationale.md.

Verify output: docs/reverse-engineering/ contains all 11 files.

If Gear 2 fails: report which files are missing. Ask user whether to retry or skip.

Proceed to BMAD Synthesize.

Gears 3-5: BMAD Synthesize

Run the bmad-synthesize skill in the selected mode (YOLO or Guided, from synthesize_mode in config).

Verify output: _bmad-output/planning-artifacts/ contains prd.md, architecture.md, epics.md, ux-design-specification.md.

If BMAD Synthesize fails: report which artifacts are missing. Ask user whether to retry or skip.

Proceed to Gear 6.

Gear 6: BMAD Handoff

If the user wants BMAD refinement: display BMAD installation instructions and *workflow-init command.

If artifacts are sufficient: mark workflow complete.

Log: "Cruise control complete. BMAD Auto-Pilot path finished."

Execution: BMAD Method Path

Gear 1: Analyze

Run the analyze skill. Set implementation_framework to bmad in .stackshift-state.json.

Verify output: analysis-report.md exists.

If Gear 1 fails: ask user whether to retry or abort.

Proceed to Gear 2.

Gear 2: Reverse Engineer

Launch the stackshift:stackshift-code-analyzer:AGENT agent via Task tool. Extract all 11 documentation files.

Verify output: docs/reverse-engineering/ contains all 11 files.

If Gear 2 fails: report which files are missing. Ask user whether to retry or skip.

Skip Gears 3-5. Proceed to Gear 6.

Gear 6: BMAD Handoff

Display instructions for BMAD installation and setup. Provide the *workflow-init command. Explain how to point BMAD to docs/reverse-engineering/.

Log: "Cruise control complete. BMAD Method path finished. BMAD takes over from here."

Execution: Architecture Only Path

Gear 1: Analyze

Run the analyze skill. Set implementation_framework to architect-only in .stackshift-state.json. Store the architecture constraints collected during setup.

Verify output: analysis-report.md exists.

If Gear 1 fails: ask user whether to retry or abort.

Proceed to Gear 2.

Gear 2: Reverse Engineer

Launch the stackshift:stackshift-code-analyzer:AGENT agent via Task tool. Extract all 11 documentation files.

Verify output: docs/reverse-engineering/ contains all 11 files.

If Gear 2 fails: report which files are missing. Ask user whether to retry or skip.

Proceed to Architecture Generator.

Gears 3-5: Architecture Generator

Run the architect skill with the user constraints from Gear 1. Generate architecture.md with Mermaid diagrams, ADRs, and infrastructure recommendations.

Verify output: architecture.md exists.

If Architecture Generator fails: report the error. Ask user whether to retry or skip.

Log: "Cruise control complete. Architecture Only path finished."

Execution: Portable Extraction Path

Gear 1: Analyze

Run the analyze skill. Set implementation_framework to portable-extract in .stackshift-state.json.

Verify output: analysis-report.md exists.

If Gear 1 fails: ask user whether to retry or abort.

Proceed to Gear 2.

Gear 2: Reverse Engineer

Launch the stackshift:stackshift-code-analyzer:AGENT agent via Task tool. Extract all 11 documentation files.

Verify output: docs/reverse-engineering/ contains all 11 files.

If Gear 2 fails: report which files are missing. Ask user whether to retry or skip.

Proceed to Portable Extraction.

Gears 3-5: Portable Extraction

Run the portable-extract skill in the selected mode (YOLO or Guided, from synthesize_mode in config).

Verify output: _portable-extract/ contains epics.md and component-spec.md.

If Portable Extraction fails: report which artifacts are missing. Ask user whether to retry or skip.

Log: "Cruise control complete. Portable Extraction path finished."

Execution: Widget Migration Path

Gear 1: Analyze

Run the analyze skill. Detect the widget type (osiris, v9-velocity, v9-viewmodel, gvm). Set implementation_framework to widget-migrate in .stackshift-state.json.

Verify output: analysis-report.md exists and widget type is identified.

If Gear 1 fails: ask user whether to retry or abort.

Skip standard Gear 2. Proceed directly to Widget Migration.

Gears 2-6: Widget Migration Pipeline

Run the widget-migrate skill in the selected mode (YOLO or Guided, from config). Pass the widget ID as an argument for argument-based invocation.

The full pipeline runs: detect, extract, preference catalog, Iris map, portable spec, targeted epics.

Verify output in docs/specs/widgets/{widget-name}/:

• preference-catalog.md
• iris-component-mapping.md
• portable-epics.md
• portable-component-spec.md
• targeted-epics.md
• migration-report.md

If Widget Migration fails: report which output files are missing. Ask user whether to retry or abort.

Log: "Cruise control complete. Widget Migration path finished."

Interruption and Resume

If the user says "stop", "pause", or "switch to manual mode":

1. Write the current gear number and completed outputs to .stackshift-state.json.
2. Set auto_mode to false in .stackshift-state.json.
3. Log: "Cruise control paused at Gear [N]. State saved. You can continue manually or resume cruise control later."

If the user says "resume cruise control":

1. Read .stackshift-state.json.
2. Set auto_mode back to true.
3. Identify the last completed gear from completedSteps.
4. Continue execution from the next gear in the path sequence.
5. Log: "Resuming cruise control from Gear [N]."

Success Criteria

GitHub Spec Kit

• .stackshift-state.json shows implementation_framework: "speckit" and all 6 gears in completedSteps
• .specify/ directory initialized with specs
• Features implemented per requested scope

BMAD Auto-Pilot

• .stackshift-state.json shows implementation_framework: "bmad-autopilot"
• docs/reverse-engineering/ with all 11 files
• _bmad-output/planning-artifacts/ with prd.md, architecture.md, epics.md, ux-design-specification.md

BMAD Method

• .stackshift-state.json shows implementation_framework: "bmad"
• docs/reverse-engineering/ with all 11 files
• BMAD handoff instructions provided

Architecture Only

• .stackshift-state.json shows implementation_framework: "architect-only"
• docs/reverse-engineering/ with all 11 files
• architecture.md generated with diagrams and ADRs

Portable Extraction

• .stackshift-state.json shows implementation_framework: "portable-extract"
• docs/reverse-engineering/ with all 11 files
• _portable-extract/epics.md and _portable-extract/component-spec.md generated

Widget Migration

• .stackshift-state.json shows implementation_framework: "widget-migrate"
• docs/specs/widgets/{widget-name}/ with all 6 output files
• Zero source-platform terms in targeted epics
• Every story references PREF-*, COMP-*, and BR-* IDs