Auto Mode

When -y or --yes: Skip clarification and confirmation prompts. Pass -y through to each step's skill invocation.

CCW

Usage

$ccw "implement user authentication with JWT"
$ccw -y "refactor the payment module"
$ccw --continue
$ccw --dry-run "add rate limiting to API endpoints"
$ccw --chain feature "add dark mode toggle"

Flags:

• -y, --yes — Auto mode: skip all prompts; propagate -y to each skill
• --continue — Resume latest paused session from last incomplete wave
• --dry-run — Display planned chain without executing
• --chain <name> — Force a specific chain (skips intent classification)

Session state: .workflow/.ccw-coordinate/{session-id}/ Core Output: tasks.csv (master) + wave-{N}-results.csv (per wave) + context.md (report)

---

Overview

Wave-based pipeline coordinator. The coordinator loop builds one wave CSV at a time, calls spawn_agents_on_csv, then performs coordinator-side artifact analysis before assembling the next wave. Barrier skills produce artifacts (plan.json, analysis results, etc.) that the coordinator reads to dynamically resolve args for subsequent steps.

Intent → Structured Extract → Resolve Chain → [Wave Loop]:
          (action×object×style)  (chainMap)
  ┌─────────────────────────────────────────────────┐
  │ 1. Identify next wave (1 or N parallel steps)   │
  │ 2. Build wave-{N}.csv with skill_call per row   │
  │ 3. spawn_agents_on_csv(wave-{N}.csv)            │
  │ 4. Read wave-{N}-results.csv                    │
  │ 5. If barrier skill: analyze artifacts,         │
  │    update context for subsequent steps           │
  │ 6. Merge into master tasks.csv                  │
  └─────────────────────────────────────────────────┘
  → Report

---

Barrier Skills

Skills that produce artifacts requiring coordinator-side analysis before the next wave can be assembled. After a barrier skill completes, the coordinator reads its output and updates the execution context.

Skill	Artifacts to Read	Context Updates
analyze-with-file	.workflow/.analysis/ANL-*/conclusions.json	analysis_dir, gaps, phase
brainstorm-with-file	.workflow/.brainstorm/*/	brainstorm_dir, features
workflow-plan	.workflow/active/WFS-*/workflow-session.json	plan_dir, task_count
workflow-lite-planex	.workflow/.lite-plan/*/plan.json	plan_dir, task_count
spec-generator	.workflow/.spec/*/	spec_session_id
roadmap-with-file	.workflow/.roadmap/*/roadmap.md	roadmap_dir
workflow-tdd-plan	.workflow/.tdd-plan/*/	tdd_plan_dir
issue-discover	.workflow/.issues/*/	issue_dir, issue_count
debug-with-file	.workflow/.debug/*/	debug_dir, findings

Non-barrier skills (can be grouped into multi-task waves): workflow-execute, workflow-test-fix-cycle, review-cycle, clean, investigate, security-audit, ship, parallel-dev-cycle, brainstorm, all team-* skills

---

Phase 1: Structured Intent Extraction

Extract a structured intent tuple using LLM semantic understanding, then route deterministically via an action × object matrix.

Extract structured intent from user input:

{
  "action":    "<from action enum>",
  "object":    "<from object enum>",
  "scope":     "<module/file/area or null>",
  "style":     "<from style enum>",
  "urgency":   "<low | normal | high>"
}

Action enum:

action	Semantic meaning
create	Build something new — feature, project, component, spec
fix	Repair something broken — fix bug, resolve error, patch
analyze	Understand deeply — analyze, investigate, discuss, explore concept
plan	Design approach — plan, break down, roadmap, decompose
execute	Implement planned work — execute, implement, develop
explore	Open-ended discovery — brainstorm, ideate, creative thinking
debug	Diagnose failures — debug, diagnose, troubleshoot
test	Run or create tests — test, generate test, TDD
review	Evaluate code quality — review, code review
refactor	Restructure code — refactor, clean up, tech debt
convert	Bridge between workflows — convert brainstorm to issue

Object enum:

object	Meaning
feature	New functionality or enhancement
bug	Defect, error, broken behavior
issue	Issue-tracker item for batch/structured management
code	Source code in general
test	Tests, test suite, test coverage
spec	Specification, PRD, product requirements
doc	Documentation
ui	User interface, design, component
performance	Performance characteristics
security	Security concerns
architecture	System architecture, design decisions
project	Entire project (greenfield)
team	Team-based execution

Style enum:

style	Meaning
quick	Fast, lightweight, minimal ceremony
documented	With file artifacts, discussion docs
collaborative	Multi-agent, multi-perspective
structured	Formal planning, spec-driven, phased
iterative	Cycle-based, self-iterating with reflection
tdd	Test-driven development
default	No specific style preference

---

Chain Map

Routing via detectTaskType(intent) → chain name → skill list.

Task Type Detection (action × object × style matrix)

function detectTaskType(intent) {
  const { action, object, style, urgency } = intent;

  // Urgency override
  if (urgency === 'high' && (action === 'fix' || object === 'bug')) return 'bugfix-hotfix';

  // Style-first routing
  if (style === 'tdd') return 'tdd';
  if (style === 'collaborative' && action === 'plan') return 'collaborative-plan';
  if (style === 'collaborative' && action === 'analyze') return 'analyze-wave';
  if (style === 'collaborative' && action !== 'plan') return 'multi-cli';
  if (style === 'iterative' && object === 'test') return 'integration-test';
  if (style === 'iterative' && action === 'refactor') return 'refactor';

  // Action × Object matrix
  const matrix = {
    'create': { 'project': 'greenfield', 'feature': 'feature', 'spec': 'spec-driven', 'test': 'test-gen', 'doc': 'documentation', 'ui': 'ui-design', 'issue': 'issue-batch', '_default': 'feature' },
    'fix':    { 'bug': 'bugfix', 'test': 'test-fix', 'issue': 'issue-batch', 'code': 'bugfix', 'security': 'bugfix', '_default': 'bugfix' },
    'analyze':{ 'architecture': 'analyze-file', 'code': 'analyze-file', 'bug': 'debug-file', 'security': 'security', '_default': 'analyze-file' },
    'explore':{ 'feature': 'brainstorm', 'architecture': 'brainstorm', 'issue': 'issue-batch', '_default': 'exploration' },
    'plan':   { 'feature': 'feature', 'project': 'greenfield', 'issue': 'issue-transition', '_default': 'feature' },
    'execute':{ 'issue': 'issue-transition', '_default': 'feature' },
    'debug':  { 'bug': style === 'documented' ? 'debug-file' : 'debug', '_default': style === 'documented' ? 'debug-file' : 'debug' },
    'test':   { 'test': 'test-fix', 'code': 'test-gen', 'feature': 'integration-test', '_default': 'test-gen' },
    'review': { '_default': 'review' },
    'refactor':{ '_default': 'refactor' },
    'convert':{ 'issue': 'brainstorm-to-issue', '_default': 'issue-transition' },
  };

  // Special compound detections
  if (action === 'plan' && style === 'structured' && /roadmap/.test(rawInput)) return 'roadmap';
  if (/csv.?wave|wave.?pipeline|并行波|波次执行/.test(rawInput)) return 'analyze-wave';
  if (object === 'team') return 'team-planex';
  if (/ship|release|publish/.test(rawInput)) return 'ship';

  const actionMap = matrix[action];
  if (!actionMap) return 'feature';
  return actionMap[object] || actionMap['_default'] || 'feature';
}

Available Skills Inventory

Skills with SKILL.md (spawn_agents_on_csv native): analyze-with-file, brainstorm, brainstorm-with-file, clean, csv-wave-pipeline, debug-with-file, issue-discover, parallel-dev-cycle, project-documentation-workflow, review-cycle, roadmap-with-file, spec-generator, workflow-execute, workflow-lite-planex, workflow-plan, workflow-tdd-plan, workflow-test-fix-cycle, team-planex, team-coordinate, team-lifecycle-v4, team-issue, team-review, team-testing, team-quality-assurance, team-tech-debt, team-perf-opt, team-arch-opt, team-brainstorm, team-ultra-analyze, team-uidesign, team-ui-polish, team-ux-improve, team-visual-a11y, team-frontend, team-frontend-debug, team-interactive-craft, team-motion-design, team-roadmap-dev

Skills with orchestrator.md (phase-based): investigate, security-audit, ship, memory-capture

Chain Definitions (task_type → skill sequence)

All $skill-name references below correspond to actual .codex/skills/{skill-name}/ directories. [B] = barrier skill (solo wave, coordinator analyzes artifacts after)

task_type	Chain name	Steps (skills, in order)
bugfix-hotfix	bugfix.hotfix	$workflow-lite-planex --hotfix
bugfix	bugfix.standard	$investigate → $workflow-lite-planex --bugfix [B] → $workflow-test-fix-cycle
feature (low)	rapid	$workflow-lite-planex [B] → $workflow-test-fix-cycle
feature (high)	coupled	$workflow-plan [B] → $workflow-execute → $review-cycle → $workflow-test-fix-cycle
greenfield	greenfield	$brainstorm-with-file [B] → $workflow-plan [B] → $workflow-execute → $workflow-test-fix-cycle
brainstorm	brainstorm-to-plan	$brainstorm-with-file [B] → $workflow-plan [B] → $workflow-execute → $workflow-test-fix-cycle
brainstorm-to-issue	brainstorm-to-issue	$brainstorm-with-file [B] → $parallel-dev-cycle
debug-file	debug-with-file	$debug-with-file
debug	investigate	$investigate
analyze-file	analyze-to-plan	$analyze-with-file [B] → $workflow-lite-planex
collaborative-plan	collaborative-plan	$brainstorm-with-file [B] → $workflow-execute
roadmap	roadmap	$roadmap-with-file [B] → $team-planex
spec-driven	spec-driven	$spec-generator [B] → $workflow-plan [B] → $workflow-execute → $workflow-test-fix-cycle
tdd	tdd	$workflow-tdd-plan [B] → $workflow-execute
test-gen	test-gen	$workflow-test-fix-cycle
test-fix	test-fix	$workflow-test-fix-cycle
review	review	$review-cycle → $workflow-test-fix-cycle
refactor	refactor	$clean
integration-test	integration-test	$workflow-test-fix-cycle
multi-cli	multi-cli	$brainstorm → $workflow-test-fix-cycle
issue-batch	issue	$issue-discover [B] → $parallel-dev-cycle
issue-transition	rapid-to-issue	$workflow-lite-planex --plan-only [B] → $parallel-dev-cycle
team-planex	team-planex	$team-planex
team-issue	team-issue	$team-issue
team-qa	team-qa	$team-quality-assurance
team-review	team-review	$team-review
team-testing	team-testing	$team-testing
documentation	docs	$project-documentation-workflow
security	security	$security-audit
ui-design	ui	$brainstorm-with-file [B] → $workflow-plan [B] → $workflow-execute
exploration	full	$brainstorm → $workflow-plan [B] → $workflow-execute → $workflow-test-fix-cycle
analyze-wave	analyze-wave	$analyze-with-file [B] → $csv-wave-pipeline → $workflow-test-fix-cycle
ship	ship	$ship

---

Implementation

Session Initialization

const dateStr = new Date().toISOString().substring(0, 10).replace(/-/g, '')
const timeStr = new Date().toISOString().substring(11, 19).replace(/:/g, '')
const sessionId = `CCW-${dateStr}-${timeStr}`
const sessionDir = `.workflow/.ccw-coordinate/${sessionId}`

Bash(`mkdir -p ${sessionDir}`)

Phase 1: Resolve Intent and Chain

--continue mode: Glob .workflow/.ccw-coordinate/CCW-*/state.json sorted by name desc; load the most recent; resume from first pending wave.

Fresh mode:

1. Read .workflow/state.json for project context (current_phase, workflow_name)
2. If --chain is given, use it directly
3. Otherwise, extract structured intent {action, object, scope, style, urgency} from user input using LLM semantic understanding
4. Route via detectTaskType(intent) matrix to get task_type
5. Assess complexity (low|medium|high) for complexity-adaptive routing
6. If no confident classification and not AUTO_YES: ask one clarifying question via AskUserQuestion
7. Resolve the chain's skill list from Chain Definitions
8. Write state.json:

Write(`${sessionDir}/state.json`, JSON.stringify({
  id: sessionId,
  intent,
  structured_intent: { action, object, scope, style, urgency },
  task_type,
  complexity,
  chain: resolvedChain,
  auto_yes: AUTO_YES,
  status: "in_progress",
  started_at: new Date().toISOString(),
  context: {
    phase: resolvedPhase,
    plan_dir: null,
    analysis_dir: null,
    brainstorm_dir: null,
    spec_session_id: null,
    roadmap_dir: null,
    tdd_plan_dir: null,
    issue_dir: null,
    debug_dir: null,
    gaps: null
  },
  waves: [],   // populated as waves execute
  steps: CHAIN_STEPS[resolvedChain].map((skill, i) => ({
    step_n: i + 1,
    skill: skill.cmd,
    args: skill.args ?? '',
    is_barrier: BARRIER_SKILLS.has(skill.cmd),
    status: "pending",
    wave_n: null
  }))
}, null, 2))

--dry-run: Display the chain plan and stop.

Chain:  <resolvedChain>
Type:   <task_type> | Complexity: <complexity>
Steps:
  1. $<cmd> <args>  [BARRIER]
  2. $<cmd> <args>
  3. $<cmd> <args>

User confirmation (skip if AUTO_YES): Display the plan above and prompt Proceed? (yes/no).

---

Phase 2: Wave Execution Loop

The coordinator iterates over pending steps, grouping them into waves and executing one wave at a time.

Wave Grouping Rules

1. A barrier skill is always alone in its wave (wave size = 1)
2. Consecutive non-barrier skills with no inter-dependencies are grouped into one wave (wave size = N)
3. After a barrier wave completes → coordinator analyzes artifacts → updates context → re-assembles subsequent step args

Per-Wave Execution

let waveNum = 0;

while (state.steps.some(s => s.status === 'pending')) {
  waveNum++;

  // 1. Determine wave contents
  const waveSteps = buildNextWave(state.steps);

  // 2. Assemble skill_call for each step (with latest context)
  const waveCsv = waveSteps.map((step, i) => ({
    id: String(step.step_n),
    skill_call: buildSkillCall(step, state.context),
    topic: `Chain "${state.chain}" step ${step.step_n}/${state.steps.length}`
  }));

  // 3. Write wave CSV
  const csvContent = 'id,skill_call,topic\n' + waveCsv.map(r =>
    `"${r.id}","${r.skill_call.replace(/"/g, '""')}","${r.topic}"`
  ).join('\n');
  Write(`${sessionDir}/wave-${waveNum}.csv`, csvContent);

  // 4. Execute wave
  spawn_agents_on_csv({
    csv_path: `${sessionDir}/wave-${waveNum}.csv`,
    id_column: "id",
    instruction: WAVE_INSTRUCTION,
    max_workers: waveSteps.length > 1 ? waveSteps.length : 1,
    max_runtime_seconds: 1800,
    output_csv_path: `${sessionDir}/wave-${waveNum}-results.csv`,
    output_schema: RESULT_SCHEMA
  });

  // 5. Read results, update step status
  const results = readCSV(`${sessionDir}/wave-${waveNum}-results.csv`);
  for (const row of results) {
    const step = state.steps.find(s => s.step_n === parseInt(row.id));
    step.status = row.status;
    step.findings = row.summary;
    step.artifacts = row.artifacts;
    step.wave_n = waveNum;
  }

  // 6. Barrier analysis (if wave contained a barrier skill)
  if (waveSteps.length === 1 && BARRIER_SKILLS.has(waveSteps[0].skill)) {
    analyzeBarrierArtifacts(waveSteps[0], results[0], state.context);
  }

  // 7. Persist state
  state.waves.push({ wave_n: waveNum, steps: waveSteps.map(s => s.step_n), results });
  Write(`${sessionDir}/state.json`, JSON.stringify(state, null, 2));

  // 8. Abort on failure
  if (results.some(r => r.status === 'failed')) {
    state.status = 'aborted';
    state.steps.filter(s => s.status === 'pending').forEach(s => s.status = 'skipped');
    Write(`${sessionDir}/state.json`, JSON.stringify(state, null, 2));
    break;
  }
}

---

Instruction Template

你是 CSV job 子 agent。

先原样执行这一段技能调用：
{skill_call}

然后基于结果完成这一行任务说明：
{topic}

限制：
- 不要修改 .workflow/.ccw-coordinate/ 下的 state 文件
- skill 内部有自己的 session 管理，按 skill SKILL.md 执行即可

最后必须调用 `report_agent_job_result`，返回 JSON：
{"status":"completed|failed","skill_call":"{skill_call}","summary":"一句话结果","artifacts":"产物路径或空字符串","error":"失败原因或空字符串"}

Result Schema

const RESULT_SCHEMA = {
  type: "object",
  properties: {
    status: { type: "string", enum: ["completed", "failed"] },
    skill_call: { type: "string" },
    summary: { type: "string" },
    artifacts: { type: "string" },
    error: { type: "string" }
  },
  required: ["status", "skill_call", "summary", "artifacts", "error"]
};

---

Barrier Analysis Logic

After a barrier skill completes, the coordinator reads its artifacts and updates state.context:

const BARRIER_SKILLS = new Set([
  'analyze-with-file', 'brainstorm-with-file', 'workflow-plan',
  'workflow-lite-planex', 'spec-generator', 'roadmap-with-file',
  'workflow-tdd-plan', 'issue-discover', 'debug-with-file'
]);

function analyzeBarrierArtifacts(step, result, ctx) {
  const artifactPath = result.artifacts;

  switch (step.skill) {
    case 'analyze-with-file':
      // Read analysis conclusions → extract gaps, phase info
      const analysisFiles = Glob('.workflow/.analysis/ANL-*/conclusions.json');
      const latest = analysisFiles.sort().pop();
      if (latest) {
        const conclusions = JSON.parse(Read(latest));
        ctx.analysis_dir = latest.replace('/conclusions.json', '');
        ctx.gaps = conclusions.gaps ?? null;
        if (!ctx.phase) ctx.phase = conclusions.phase ?? null;
      }
      break;

    case 'brainstorm-with-file':
      ctx.brainstorm_dir = artifactPath;
      break;

    case 'workflow-plan':
      // Read workflow session → know task structure for execute
      const wfSessions = Glob('.workflow/active/WFS-*/workflow-session.json');
      const latestWf = wfSessions.sort().pop();
      if (latestWf) {
        const session = JSON.parse(Read(latestWf));
        ctx.plan_dir = latestWf.replace('/workflow-session.json', '');
        ctx.task_count = session.tasks?.length ?? 0;
      }
      break;

    case 'workflow-lite-planex':
      const litePlans = Glob('.workflow/.lite-plan/*/plan.json');
      const latestPlan = litePlans.sort().pop();
      if (latestPlan) {
        ctx.plan_dir = latestPlan.replace('/plan.json', '');
        ctx.task_count = JSON.parse(Read(latestPlan)).tasks?.length ?? 0;
      }
      break;

    case 'spec-generator':
      ctx.spec_session_id = artifactPath;
      break;

    case 'roadmap-with-file':
      ctx.roadmap_dir = artifactPath;
      break;

    case 'workflow-tdd-plan':
      ctx.tdd_plan_dir = artifactPath;
      break;

    case 'issue-discover':
      ctx.issue_dir = artifactPath;
      break;

    case 'debug-with-file':
      ctx.debug_dir = artifactPath;
      ctx.findings = result.summary;
      break;
  }
}

Skill Call Assembly

The coordinator builds each skill_call with resolved context — sub-agents just execute verbatim:

const AUTO_FLAG_MAP = {
  'brainstorm-with-file': '-y',
  'analyze-with-file': '-y',
  'debug-with-file': '-y',
  'workflow-plan': '-y',
  'workflow-lite-planex': '-y',
  'workflow-execute': '-y',
  'workflow-test-fix-cycle': '-y',
  'workflow-tdd-plan': '-y',
  'spec-generator': '-y',
  'roadmap-with-file': '-y',
  'issue-discover': '-y',
  'parallel-dev-cycle': '-y',
  'review-cycle': '-y',
  'clean': '-y',
  'brainstorm': '-y',
  'csv-wave-pipeline': '-y',
};

function buildSkillCall(step, ctx) {
  let args = (step.args ?? '')
    .replace(/{intent}/g, state.intent ?? '')
    .replace(/{phase}/g, ctx.phase ?? '')
    .replace(/{plan_dir}/g, ctx.plan_dir ?? '')
    .replace(/{analysis_dir}/g, ctx.analysis_dir ?? '')
    .replace(/{brainstorm_dir}/g, ctx.brainstorm_dir ?? '')
    .replace(/{spec_session_id}/g, ctx.spec_session_id ?? '')
    .replace(/{roadmap_dir}/g, ctx.roadmap_dir ?? '')
    .replace(/{tdd_plan_dir}/g, ctx.tdd_plan_dir ?? '')
    .replace(/{issue_dir}/g, ctx.issue_dir ?? '')
    .replace(/{debug_dir}/g, ctx.debug_dir ?? '');

  // Inject intent as first arg if no args present
  if (!args.trim()) args = `"${state.intent}"`;

  if (state.auto_yes) {
    const flag = AUTO_FLAG_MAP[step.skill];
    if (flag && !args.includes(flag)) args = args ? `${args} ${flag}` : flag;
  }

  return `$${step.skill} ${args}`.trim();
}

function buildNextWave(steps) {
  const pending = steps.filter(s => s.status === 'pending');
  if (!pending.length) return [];

  const first = pending[0];
  // Barrier skill → solo wave
  if (BARRIER_SKILLS.has(first.skill)) return [first];

  // Group consecutive non-barriers
  const wave = [first];
  for (let i = 1; i < pending.length; i++) {
    if (BARRIER_SKILLS.has(pending[i].skill)) break;
    wave.push(pending[i]);
  }
  return wave;
}

---

Phase 3: Completion Report

state.status = state.steps.every(s => s.status === 'completed') ? 'completed' : state.status;
state.completed_at = new Date().toISOString();
Write(`${sessionDir}/state.json`, JSON.stringify(state, null, 2));

Generate context.md:

# CCW Coordinate Report — {chain}

## Summary
- Session: {sessionId}
- Chain: {chain}
- Type: {task_type} | Complexity: {complexity}
- Waves: {waveNum} executed
- Steps: {completed}/{total} completed

## Wave Results
### Wave {N} (barrier: {skill})
| Step | Skill Call | Status | Summary |
|------|-----------|--------|---------|
| {step_n} | {skill_call} | {status} | {summary} |

Artifacts: {artifacts}
Context update: {what changed}

Display:

=== CCW COORDINATE COMPLETE ===
Session:  <sessionId>
Chain:    <chain>
Type:     <task_type> | Complexity: <complexity>
Waves:    <N> executed
Steps:    <completed>/<total>

WAVE RESULTS:
  [W1] $analyze-with-file -y       →  ✓  found 3 gaps       [BARRIER]
  [W2] $workflow-lite-planex -y     →  ✓  12 tasks planned   [BARRIER]
  [W3] $workflow-test-fix-cycle -y  →  ✓  all tests pass

State:    .workflow/.ccw-coordinate/<sessionId>/state.json
Resume:   $ccw --continue

---

CSV Schema

wave-{N}.csv (Per-Wave Input)

id,skill_call,topic
"1","$analyze-with-file ""fix auth"" -y","Chain ""bugfix.standard"" step 1/3"

Column	Description
id	Step number from chain (string)
skill_call	Full skill invocation assembled by coordinator with resolved context
topic	Brief description for the agent

tasks.csv (Master State)

id,skill,args,wave_n,status,findings,artifacts,error

Accumulated across all waves. Updated after each wave completes.

---

Error Handling

Code	Severity	Condition	Recovery
E001	error	Intent unclassifiable after clarification	Default to feature chain (rapid)
E002	error	--chain value not in chain map	List valid chains, abort
E003	error	Wave timeout (max_runtime_seconds)	Mark step failed, abort chain
E004	error	Barrier artifact not found	Retry wave once, then abort
E005	error	--continue: no session found	List sessions, prompt
W001	warning	Barrier artifact partial	Continue with available context

---

Core Rules

1. Start Immediately: Init session dir and write state.json before any wave
2. Wave-by-wave: Never start wave N+1 before wave N results are read and analyzed
3. Barrier = solo wave: A barrier skill always executes alone; coordinator analyzes its artifacts before proceeding
4. Non-barriers can parallel: Consecutive non-barrier skills in the same wave execute with max_workers = N
5. Coordinator owns context: Sub-agents never read prior results — coordinator assembles the full skill_call with resolved args
6. Simple instruction: Sub-agent instruction is minimal — just "execute {skill_call}, report result"
7. Abort on failure: Failed step → mark remaining as skipped → report
8. State.json tracks waves: Each wave is recorded with step IDs and results for resume
9. Dry-run is read-only: Display chain with [BARRIER] markers, no execution
10. Resume from wave: --continue finds last completed wave and resumes from next pending step
11. Semantic Routing: Use LLM structured extraction (action × object × style) not regex for intent classification