RCP-CWK-035 Cowork CPM Phase Runner

Executes a single phase within a CPM initiative by routing to the appropriate execution engine

Handles the full lifecycle of phase execution: setup, delegation, artifact collection, verification, and completion reporting. Returns a structured completion report to CWK-034 (CPM Orchestrator) for phase transition processing.

Cowork CPM Phase Runner

Cowork | CPM | Difficulty: Advanced | Time: 5–30 min per phase
Tags: Introduced in Beta, CPM, Phase Execution, Sub-Agent, CV Workflow, Verification, Artifacts, Cowork

TL;DR

What It Does
Executes a single phase within a CPM initiative by routing to the appropriate execution engine. Handles the full lifecycle of phase execution: setup, delegation, artifact collection, verification, and completion reporting. Returns a structured completion report to CWK-034 (CPM Orchestrator) for phase transition processing.
How It Works
Four execution modes. CV_WORKFLOW delegates to CWK-030 for full Creator/Validator quality loops — best for content that needs iterative refinement. SUB_AGENT delegates to CWK-008 with one of four delegation patterns (parallel research, sequential handoff, builder/validator, single delegate). DIRECT executes inline without sub-agents — the parent AI performs the work from a provided prompt. CUSTOM reads and executes a prompt from a file. After execution, artifacts are collected and matched against expected deliverables, then verified through CWK-024 (Verification Gate) with confidence scoring. A structured completion report is generated with status (PASSED, PASSED_WITH_NOTES, or REVISION_NEEDED), artifact inventory, and verification confidence.
What To Expect
A completion report JSON file in the phase output directory. All artifacts written to {output_dir}/artifacts/. An artifact inventory mapping deliverables to files. The completion report is structured for direct consumption by CWK-034’s transition mode — pass it as the completion_report parameter.
Best Results When You
Choose the right execution mode for your phase. Use CV_WORKFLOW for content requiring quality iteration. Use SUB_AGENT for parallelizable research or multi-step workflows. Use DIRECT for simple, well-defined tasks. Define deliverables precisely so artifact collection can verify completeness.
Typical Time
CV_WORKFLOW: 10–30 minutes (inherits CWK-030 timing). SUB_AGENT: 5–20 minutes depending on delegation pattern and task count. DIRECT: 2–10 minutes. CUSTOM: varies by prompt complexity. Verification adds 1–2 minutes.
Difficulty
Advanced — requires understanding of the four execution modes, their parameter requirements, and how to structure deliverables for accurate artifact collection. Typically called by CWK-034 rather than directly by users.

How To Start

This recipe executes a single CPM phase. It is typically called by CWK-034 (CPM Orchestrator) but can be used standalone. Choose the execution mode that matches your phase’s needs.

STEP 1Define Phase and Execution Mode

Provide the phase identity, execution mode, and mode-specific parameters.
Available parameters
  • phase_id · string · required
    Phase identifier matching the CPM state.
  • execution_mode · string · required
    How to execute this phase: “CV_WORKFLOW”, “SUB_AGENT”, “DIRECT”, or “CUSTOM”.
  • target_project_path · string · required
    Path to the child project workspace.
  • deliverables · list · required
    Expected deliverable names for artifact verification.
  • scope_description · string · required
    What work this phase must accomplish.
  • success_criteria · string · required
    How to verify phase completion.
  • content_type · string · required for CV_WORKFLOW
    Content type passed to CWK-030.
  • content_name · string · required for CV_WORKFLOW
    Deliverable name passed to CWK-030.
  • max_cv_cycles · integer · optional · default: 3
    Maximum Creator/Validator cycles for CV_WORKFLOW.
  • delegation_pattern · string · required for SUB_AGENT
    CWK-008 delegation pattern: “parallel_research”, “sequential_handoff”, “builder_validator”, or “single_delegate”.
  • direct_prompt · string · required for DIRECT
    Prompt for inline execution.
  • custom_prompt_file · string · required for CUSTOM
    Path to file containing the custom prompt.
  • output_dir · string · optional
    Phase output directory. Defaults to {target_project_path}/.cpmPhase-{phase_id}/.
  • enable_git_checkpoint · boolean · optional · default: true
    Call CWK-005 after completion.
  • phase_context · dict · optional · default: {}
    Extra context including prior phase artifacts and references.
Invocations
Execute via CV Workflow
#H->AI::Directive: (Execute CWK-035 with phase_id="Phase 1", execution_mode="CV_WORKFLOW", content_type="documentation", content_name="API Guide", …)
Execute via sub-agent delegation
#H->AI::Directive: (Execute CWK-035 with phase_id="Phase 2", execution_mode="SUB_AGENT", delegation_pattern="parallel_research", …)
Execute directly
#H->AI::Directive: (Execute CWK-035 with phase_id="Phase 3", execution_mode="DIRECT", direct_prompt="…", …)

Usage Examples

EXAMPLE 1CV_WORKFLOW phase execution

User
Execute Phase 1 using the Creator/Validator workflow.
Cat
Phase Runner pre-flight for Phase 1. Validating parameters. Delegating to CWK-030 (CV Workflow). Auguste creates API Guide draft. Fernand validates. Cycle 1 complete — confidence 82%. Collecting artifacts: 2/2 deliverables present. Verification gate: 85% confidence, 4/4 gates passed. Status: PASSED. Completion report written. Next: Return report to CWK-034 with mode=transition.

EXAMPLE 2SUB_AGENT phase execution

User
Execute Phase 2 with parallel research delegation.
Cat
Phase Runner pre-flight for Phase 2. Delegating to CWK-008 (parallel_research). 3 sub-agents spawned for 3 deliverables. All agents completed. Collecting artifacts: 3/3 deliverables present. Verification gate: 78% confidence, 3/4 gates passed. Status: PASSED_WITH_NOTES. Note: Gate 4 (cross-reference consistency) flagged minor issues. Completion report written.

EXAMPLE 3DIRECT phase execution

User
Execute Phase 3 directly.
Cat
Phase Runner pre-flight for Phase 3. Executing directly (no sub-agents). Work complete. Collecting artifacts: 1/1 deliverables present. Verification gate: 92% confidence, 4/4 gates passed. Status: PASSED. Completion report written. Git checkpoint saved.

How AI Reads This Recipe

When this recipe is triggered, the AI executes a CPM phase. The AI should:
  1. Pre-Flight (Step 0). Validate phase_id and target_project_path. Load CPM state if .cpmrc.json exists and verify phase status is READY or IN_PROGRESS. Validate execution_mode and its mode-specific required parameters. Set up output directory. Collect prior phase artifacts from phase_context.
  2. Mode-Specific Setup (Step 1). Build parameters for the execution engine. CV_WORKFLOW: map phase definition to CWK-030 parameters including scope as requirements, success criteria, prior artifacts as research files. SUB_AGENT: map deliverables to CWK-008 delegation tasks with self-contained context block (sub-agents do not read CLAUDE.md). DIRECT: render prompt with phase variable substitution. CUSTOM: read prompt file, validate self-containment.
  3. Execute (Step 2). Update CPM state to IN_PROGRESS. Dispatch to the execution engine. CV_WORKFLOW calls CWK-030. SUB_AGENT calls CWK-008. DIRECT and CUSTOM execute inline.
  4. Collect Artifacts (Step 3). Scan output directory for files. Match against expected deliverables. Flag missing items. Write artifact inventory JSON. Determine preliminary status: all present → PASSED, partial → PASSED_WITH_NOTES, none → REVISION_NEEDED.
  5. Verification Gate (Step 4). Build verification claims from deliverables and success criteria. Call CWK-024 with claims and evidence files. Adjust phase status based on confidence score: above threshold → keep status, 50–threshold → PASSED_WITH_NOTES, below 50 → REVISION_NEEDED.
  6. Completion Report (Step 5). Build structured report with phase_id, status, execution details, artifact list, missing deliverables, verification confidence, and notes. Write to completion-report.json. Display summary. Instruct user to pass report to CWK-034 transition mode.
  7. Git Checkpoint (Step 6). If enabled, call CWK-005 with phase completion message.

When To Use This Recipe

Use this recipe when you need to execute a phase within a CPM initiative managed by CWK-034, want structured artifact collection and verification for a defined set of deliverables, or need to route phase execution to different engines based on the work type (quality content → CV_WORKFLOW, research → SUB_AGENT, simple tasks → DIRECT).
Do Not Use When
You are not working within a CPM initiative — use CWK-030 (Creator/Validator) or CWK-008 (Sub-Agent Delegation) directly. For multi-persona collaboration without cross-project context, use CWK-031. This recipe is specifically designed as the execution layer for CWK-034.

Recipe FAQ

What is the difference between CWK-035 and CWK-030?
CWK-035 is a phase execution wrapper that can call CWK-030 as one of its four execution engines. CWK-035 adds artifact collection, CWK-024 verification, and structured completion reporting on top of whatever execution engine it uses. CWK-030 runs a Creator/Validator quality loop; CWK-035 wraps that in CPM lifecycle context.
Can I use CWK-035 without CWK-034?
Yes — it works standalone. However, the completion report is structured for CWK-034 consumption. Without CWK-034, you would manually process the completion report rather than feeding it into an initiative transition.
What are the four delegation patterns for SUB_AGENT mode?
parallel_research (multiple agents research simultaneously), sequential_handoff (agents pass work forward in sequence), builder_validator (one agent builds, another validates), single_delegate (one agent handles everything). These map to CWK-008’s delegation patterns.
How does artifact collection work?
The recipe scans the output directory for files and matches them against your declared deliverables list by filename. If a deliverable name appears in an artifact filename, it is matched. Missing deliverables are flagged and affect the phase status.
What determines the final phase status?
Two factors: artifact completeness (all deliverables present vs missing) and CWK-024 verification confidence. If all deliverables are present and confidence exceeds the threshold, status is PASSED. Missing deliverables or low confidence can downgrade to PASSED_WITH_NOTES or REVISION_NEEDED.

Version History

v1.00a — 2026-03-03
Initial creation. Replaces Studio STU-061 (CPM Phase Prompt Generator) with autonomous execution. Four execution modes (CV_WORKFLOW, SUB_AGENT, DIRECT, CUSTOM), structured artifact collection and matching, CWK-024 verification gate integration, completion report generation for CWK-034 consumption, and CWK-005 git checkpoint integration.

Get this recipe with CRAFT for Claude Cowork

Cowork recipes ship bundled with CRAFT for Claude Cowork — there’s no separate download. Clone the framework once, and your AI runs every recipe automatically when invoked.

Pull anytime to stay on the latest version — free to clone, no login or email required.

Then start your session

Once CRAFT is in your project folder, open a new Cowork session and ask Claude to initialize. For example:

You

Please initialize my CRAFT session.

Claude

CRAFT session ready. Your project is loaded, your persona is active, and your recipes are available. What would you like to work on?

What is CRAFT for Claude Cowork?

Not familiar with Git? Download as a ZIP

No command line needed. Just download, move, and unzip:

  1. Open the CRAFT framework repo on GitHub.
  2. Click the green Code button, then choose Download ZIP.
  3. Move the downloaded ZIP into your Claude Cowork project folder.
  4. Unzip it: double-click on Mac, or right-click → Extract All on Windows.

Similar Posts