cc-p4p-router

cc-p4p Router

EXECUTION ENGINE. When loaded: Detect intent -> Load memory -> Execute workflow -> Update memory.

NEVER list capabilities. ALWAYS execute.

Decision Tree (FOLLOW IN ORDER)

Priority	Signal	Keywords	Workflow
1	SPEC	spec, prd, feature, requirement, user story, acceptance criteria, "write a spec", "define requirements"	SPEC
2	RESEARCH	research, interview, feedback, survey, competitive, persona, insight, "synthesize", "analyze interviews"	RESEARCH
3	ROADMAP	roadmap, priority, plan, timeline, rice, moscow, okr, quarter, "prioritize", "what should we build next"	ROADMAP
4	COMMUNICATE	update, status, stakeholder, announcement, launch, report, "send update", "draft email"	COMMUNICATE
5	METRICS	metric, kpi, dashboard, retention, conversion, north star, "define metrics", "track success"	METRICS
6	DEFAULT	Everything else	SPEC

Conflict Resolution: SPEC signals win over ROADMAP. "Plan the feature" = SPEC (not ROADMAP). "Plan the quarter" = ROADMAP.

Disambiguation: If intent is genuinely unclear, use AskUserQuestion: "What would you like to do?"

• A. Write a feature spec/PRD
• B. Synthesize research findings
• C. Plan/prioritize a roadmap
• D. Draft a stakeholder communication

Agent Chains

Workflow	Agent	Type
SPEC	spec-writer	WRITE
RESEARCH	research-synthesizer	WRITE
ROADMAP	roadmap-planner	WRITE
COMMUNICATE	comms-drafter	READ-ONLY
METRICS	Router inline (no agent)	INLINE

Each workflow = single agent + memory update. No parallel chains (v1 simplicity).

Memory (PERMISSION-FREE)

LOAD FIRST (Before routing):

Step 1 - Create directory (MUST complete before Step 2):

Bash(command="mkdir -p .claude/cc-p4p")

Step 2 - Load memory files (AFTER Step 1 completes):

Read(file_path=".claude/cc-p4p/activeContext.md")
Read(file_path=".claude/cc-p4p/insights.md")
Read(file_path=".claude/cc-p4p/roadmap-state.md")

IMPORTANT: Do NOT run Step 1 and Step 2 in parallel. Wait for mkdir to complete before reading files.

If any memory file is missing:

• Create it with Write(...) using the templates from cc-p4p:pm-memory (include the contract comment + required headings).
• Then Read(...) it before continuing.

TEMPLATE VALIDATION GATE (Auto-Heal):

After loading memory files, ensure ALL required sections exist.

activeContext.md - Required Sections

## Current Focus, ## Recent Work, ## Decisions, ## Open Questions, ## References, ## Last Updated

insights.md - Required Sections

## User Research Insights, ## Competitive Intelligence, ## Customer Feedback Themes, ## Personas, ## Last Updated

roadmap-state.md - Required Sections

## Current Priorities, ## Shipped, ## Dependencies, ## Capacity Notes, ## Last Updated

Auto-heal pattern (apply to EACH file, EACH missing section):

activeContext.md example:

Edit(file_path=".claude/cc-p4p/activeContext.md",
     old_string="## Last Updated",
     new_string="## Open Questions\n- [None yet]\n\n## Last Updated")
Read(file_path=".claude/cc-p4p/activeContext.md")

insights.md example:

Edit(file_path=".claude/cc-p4p/insights.md",
     old_string="## Last Updated",
     new_string="## Personas\n- [None identified yet]\n\n## Last Updated")
Read(file_path=".claude/cc-p4p/insights.md")

roadmap-state.md example:

Edit(file_path=".claude/cc-p4p/roadmap-state.md",
     old_string="## Last Updated",
     new_string="## Capacity Notes\n- [Not yet assessed]\n\n## Last Updated")
Read(file_path=".claude/cc-p4p/roadmap-state.md")

VERIFY: After ALL auto-heals, Read each file once more and confirm every required section exists.

UPDATE (Final):

• Do a workflow-final memory update after the agent completes.
• Use Edit tool on memory files (permission-free), then Read-back verify.

Memory update rules:

1. Use Edit(...) (not Write) to update existing .claude/cc-p4p/*.md.
2. Immediately Read(...) the edited file and confirm the expected text exists.
3. If the update did not apply, STOP and retry with a correct, exact old_string anchor.

Check Active Workflow Tasks

After loading memory, check for active tasks:

TaskList()  # Check for pending/in-progress workflow tasks

If active CC-P4P workflow task exists (subject starts with CC-P4P ):

• Resume from task state (use TaskGet({ taskId }))
• Skip workflow selection - continue execution from where it stopped
• Check blockedBy to determine which agent to run next

If no active tasks:

• Proceed with workflow selection below

---

Task-Based Orchestration

At workflow start, create task hierarchy using TaskCreate/TaskUpdate:

SPEC Workflow Tasks

TaskCreate({
  subject: "CC-P4P SPEC: {feature_summary}",
  description: "User request: {request}\n\nWorkflow: SPEC\nAgent: spec-writer",
  activeForm: "Writing spec for {feature}"
})
# Returns workflow_task_id

TaskCreate({
  subject: "CC-P4P spec-writer: Write spec for {feature}",
  description: "Write feature spec/PRD per user request",
  activeForm: "Writing feature spec"
})
# Returns agent_task_id

TaskCreate({
  subject: "CC-P4P Memory Update: Persist spec learnings",
  description: "REQUIRED: Collect Memory Notes from spec-writer and persist to memory files.\n\n**Instructions:**\n1. Find '### Memory Notes' section from agent output\n2. Update .claude/cc-p4p/activeContext.md with spec reference\n3. Use Read-Edit-Read pattern for each file.",
  activeForm: "Persisting spec learnings"
})
# Returns memory_task_id
TaskUpdate({ taskId: memory_task_id, addBlockedBy: [agent_task_id] })

RESEARCH Workflow Tasks

TaskCreate({
  subject: "CC-P4P RESEARCH: {topic_summary}",
  description: "User request: {request}\n\nWorkflow: RESEARCH\nAgent: research-synthesizer",
  activeForm: "Synthesizing research on {topic}"
})

TaskCreate({
  subject: "CC-P4P research-synthesizer: Synthesize {topic}",
  description: "Synthesize research findings per user request",
  activeForm: "Synthesizing research"
})
# Returns agent_task_id

TaskCreate({
  subject: "CC-P4P Memory Update: Persist research insights",
  description: "REQUIRED: Persist insights to .claude/cc-p4p/insights.md.\n\nUse Read-Edit-Read pattern.",
  activeForm: "Persisting research insights"
})
# Returns memory_task_id
TaskUpdate({ taskId: memory_task_id, addBlockedBy: [agent_task_id] })

ROADMAP Workflow Tasks

TaskCreate({
  subject: "CC-P4P ROADMAP: {scope_summary}",
  description: "User request: {request}\n\nWorkflow: ROADMAP\nAgent: roadmap-planner",
  activeForm: "Planning roadmap for {scope}"
})

TaskCreate({
  subject: "CC-P4P roadmap-planner: Plan {scope}",
  description: "Plan/prioritize roadmap per user request",
  activeForm: "Planning roadmap"
})
# Returns agent_task_id

TaskCreate({
  subject: "CC-P4P Memory Update: Persist roadmap state",
  description: "REQUIRED: Persist to .claude/cc-p4p/roadmap-state.md.\n\nUse Read-Edit-Read pattern.",
  activeForm: "Persisting roadmap state"
})
# Returns memory_task_id
TaskUpdate({ taskId: memory_task_id, addBlockedBy: [agent_task_id] })

COMMUNICATE Workflow Tasks

TaskCreate({
  subject: "CC-P4P COMMUNICATE: {comm_summary}",
  description: "User request: {request}\n\nWorkflow: COMMUNICATE\nAgent: comms-drafter (READ-ONLY)",
  activeForm: "Drafting communication for {audience}"
})

TaskCreate({
  subject: "CC-P4P comms-drafter: Draft {comm_type}",
  description: "Draft communication. Agent is READ-ONLY -- outputs draft text.",
  activeForm: "Drafting communication"
})
# Returns agent_task_id

TaskCreate({
  subject: "CC-P4P Memory Update: Log communication",
  description: "REQUIRED: Log communication in activeContext.md ## Recent Work.\n\nNOTE: comms-drafter is READ-ONLY. Router persists if user approves draft.",
  activeForm: "Logging communication"
})
# Returns memory_task_id
TaskUpdate({ taskId: memory_task_id, addBlockedBy: [agent_task_id] })

METRICS Workflow (Inline - No Agent)

TaskCreate({
  subject: "CC-P4P METRICS: {metric_summary}",
  description: "User request: {request}\n\nWorkflow: METRICS (inline)\nRouter handles directly using pm-roadmap-frameworks + pm-verification knowledge.",
  activeForm: "Defining metrics for {area}"
})
# Router handles this directly -- no agent spawned

Workflow Execution

SPEC

1. Load memory -> Check for existing specs in References
2. Clarify scope -> AskUserQuestion if vague
3. Create task hierarchy
4. Invoke agent:

   Task(subagent_type="cc-p4p:spec-writer", prompt="
   ## Task Context
   - **Task ID:** {taskId}
   
   ## User Request
   {request}
   
   ## Memory Summary
   {brief summary from activeContext.md}
   
   ## Existing Context
   {relevant insights from insights.md}
   
   ## SKILL_HINTS (conditional skills if triggered)
   {detected skills from CLAUDE.md Complementary Skills table, otherwise 'None'}
   
   ---
   IMPORTANT:
   - **NEVER call `EnterPlanMode`.** This is an execution agent.
   - Update `.claude/cc-p4p/{activeContext,insights,roadmap-state}.md` at the end per `cc-p4p:pm-memory`.
   - Include Router Contract in output.
   
   Execute the task and include 'Task {TASK_ID}: COMPLETED' in your output when done.
   ")
   

5. Validate Router Contract
6. Update memory (task-enforced)

RESEARCH

1. Load memory -> Check insights.md for related themes
2. Clarify scope -> AskUserQuestion: Users? Market? Competitive? All?
3. Create task hierarchy
4. Invoke research-synthesizer agent (same pattern as SPEC)
5. Validate Router Contract
6. Update memory (persist to insights.md)

ROADMAP

1. Load memory -> Load roadmap-state.md (current priorities)
2. Clarify format -> AskUserQuestion: Now/Next/Later? RICE scoring? Quarterly?
3. Create task hierarchy
4. Invoke roadmap-planner agent
5. Validate Router Contract
6. Update memory (persist to roadmap-state.md)

COMMUNICATE

1. Load memory -> Get full context from all 3 files
2. Clarify audience + cadence -> AskUserQuestion: Executive? Engineering? Customer? Weekly/monthly?
3. Create task hierarchy
4. Invoke comms-drafter agent (READ-ONLY)
5. Present draft to user
6. If user approves: Router persists to file and updates memory
7. If user requests changes: Re-invoke with feedback

METRICS (Inline)

1. Load memory
2. Identify metric area from request
3. Apply PM knowledge directly:
   • North Star metric definition
   • Leading vs lagging indicators
   • Measurement methods
   • Dashboard recommendations
4. Output metric definitions
5. Update memory with metric decisions

Router Contract Validation

After agent completes, check for Router Contract:

Look for "### Router Contract (MACHINE-READABLE)" section in agent output.
If found -> Parse and validate.
If NOT found -> Non-compliant. Ask user to review output manually.

Parse and Validate Contract:

CONTRACT FIELDS:
- STATUS: Agent's self-reported status
- CONFIDENCE: 0-100 score
- BLOCKING: true/false
- MEMORY_NOTES: Structured notes for persistence

VALIDATION:
1. If contract.BLOCKING == true:
   -> Present issues to user via AskUserQuestion
   -> Options: Fix / Skip / Abort
2. Collect contract.MEMORY_NOTES for workflow-final persistence
3. If none of above triggered -> Workflow complete

Agent Invocation Template

Pass task ID, context, and conditional skills to each agent:

Task(subagent_type="cc-p4p:{agent_name}", prompt="
## Task Context
- **Task ID:** {taskId}

## User Request
{request}

## Memory Summary
{brief summary from activeContext.md}

## PM Context
{relevant context from insights.md and roadmap-state.md}

## SKILL_HINTS (conditional skills if triggered)
{detected skills from CLAUDE.md Complementary Skills table, otherwise 'None'}
**If skills listed:** Call `Skill(skill="{skill-name}")` after memory load. If unavailable, note in Memory Notes and continue.

---
IMPORTANT:
- **NEVER call `EnterPlanMode`.** This is an execution agent.
- If your tools include `Edit`, update `.claude/cc-p4p/*.md` at the end.
- If your tools do NOT include `Edit`, include a `### Memory Notes` section.
- Include Router Contract in your output.

Execute the task and include 'Task {TASK_ID}: COMPLETED' in your output when done.
")

Workflow-Final Memory Persistence (Task-Enforced)

When the Memory Update task becomes available:

1. Review agent output for ### Memory Notes section
2. Follow the task description to persist learnings
3. Use Read-Edit-Read pattern for each memory file
4. Mark task completed

Why task-enforced:

• Tasks survive context compaction
• Visible in TaskList() -- can't be forgotten
• Workflow isn't complete until Memory Update task is done

Gates (Must Pass)

1. MEMORY_LOADED - Before routing
2. TASKS_CHECKED - Check TaskList() for active workflow
3. INTENT_CLARIFIED - User intent is unambiguous
4. TASKS_CREATED - Workflow task hierarchy created
5. ALL_TASKS_COMPLETED - All workflow tasks completed
6. MEMORY_UPDATED - Before marking done