M Agentic Feature

---
description: Execute new feature development workflow with TDD and pair programming
argument-hint: [feature description]
allowed-tools: Read(*), Edit(*), Write(*), Bash(*), Grep(*), Glob(*), TodoWrite(*), Task(*)
---

You received the command /m_agentic_feature. This command orchestrates a sophisticated TDD + pair programming workflow for feature development following Memory Bank standards.

**Your task**: $ARGUMENTS

## CRITICAL: You Are The Orchestrator

You do NOT execute the task yourself. You orchestrate specialized agents to do the work through a structured TDD + pair programming process.

## Workflow Overview

PHASE 1: Memory Bank Preparation PHASE 2A: Select Agent-A (Acceptance Criteria Definer) PHASE 2B: Select Agent-B (Task Executor) PHASE 3: TDD + Pair Programming-like Execution Loop (max 20 iterations) PHASE 4: Memory Bank Updates

---

## PHASE 1: Memory Bank Preparation

1. Find the corresponding specification in `.memory_bank/specs/`. If it doesn't exist, request the path from the user.
2. Carefully study `.memory_bank/workflows/new_feature.md`
3. Study `.memory_bank/tech_stack.md` for approved libraries and prohibited practices
4. Prepare full context for agent selection

---

## PHASE 2A: Select Agent-A (Acceptance Criteria Definer)

Run TDD-style agent selection loop with hiring-manager and hr-agent:

Loop until hiring-manager approves:

Step 1: Use Task tool with hiring-manager Input: Task description + context from Phase 1 Request: "Create comprehensive agent specification for an agent that will: - Define acceptance criteria for this task - Determine task kind (coding/design/documentation/etc) - For coding: Define test suite - For other: Define requirements checklist - Review implementations against criteria - Have veto power (but be reasonable)" Output: Agent specification document

Step 2: Use Task tool with hr-agent Input: Agent specification from Step 1 Request: "Match existing agent OR create new agent following this spec" Output: Agent name (e.g., "acceptance-criteria-definer")

Step 3: Use Task tool with hiring-manager Input: Agent name + path to .claude/agents/.md Request: "Review this agent's prompt file. Validate: - Frontmatter is correct - Has sufficient permissions - Matches requirements from specification - No scope creep Approve OR provide specific fix requests" Output: APPROVED or list of fix requests

Step 4 (if not approved): Use Task tool with hr-agent Input: Fix requests from Step 3 Request: "Apply these fixes to the agent" Output: Confirmation of fixes applied

Loop back to Step 3 until APPROVED

**Store Agent-A name** for Phase 3.

---

## PHASE 2B: Select Agent-B (Task Executor)

Repeat the same TDD-style selection loop for Agent-B:

Loop until hiring-manager approves:

Step 1: Use Task tool with hiring-manager Input: Task description + context from Phase 1 Request: "Create comprehensive agent specification for an agent that will: - Execute the task implementation - Follow TDD process (write test → implement → verify) - Work with acceptance criteria from Agent-A - Iterate based on Agent-A's reviews" Output: Agent specification document

Step 2: Use Task tool with hr-agent Input: Agent specification from Step 1 Request: "Match existing agent OR create new agent following this spec" Output: Agent name (e.g., "feature-implementer")

Step 3: Use Task tool with hiring-manager Input: Agent name + path to .claude/agents/.md Request: "Review this agent's prompt file. Validate: - Frontmatter is correct - Has sufficient permissions (likely Read, Write, Edit, Bash, Grep, Glob) - Matches requirements from specification - No scope creep Approve OR provide specific fix requests" Output: APPROVED or list of fix requests

Step 4 (if not approved): Use Task tool with hr-agent Input: Fix requests from Step 3 Request: "Apply these fixes to the agent" Output: Confirmation of fixes applied

Loop back to Step 3 until APPROVED

**Store Agent-B name** for Phase 3.

---

## PHASE 3: TDD Execution Loop (Max 20 Iterations)

For iteration 1 to 20:

Step 1: Use Task tool with Agent-A Input: - Task description - Current iteration number - Acceptance criteria file path: .memory_bank/specs/_acceptance_criteria.md - Previous implementation results (if iteration > 1)

Request: "
  {IF iteration == 1}
    Define comprehensive acceptance criteria for this task.
    Determine task kind (coding/design/documentation/etc).
    For coding: Define test suite requirements.
    For other: Define requirements checklist.
    Save criteria to: .memory_bank/specs/<task-slug>_acceptance_criteria.md
  {ELSE}
    Review Agent-B's implementation from iteration {iteration-1}.
    Check against acceptance criteria in the file.
    Provide detailed feedback:
      - What passed
      - What failed
      - Specific improvements needed
    Decision: APPROVED (all criteria met) OR REJECTED (needs work)
    Be reasonable in your veto power - don't reject for trivial issues.
  {ENDIF}
"

Output:
  - Iteration 1: Acceptance criteria file created
  - Iteration >1: Review decision (APPROVED/REJECTED) + feedback

IF APPROVED: Exit loop with success

Step 2: Use Task tool with Agent-B Input: - Task description - Acceptance criteria file: .memory_bank/specs/_acceptance_criteria.md - Feedback from Agent-A (if iteration > 1) - Spec file from .memory_bank/specs/ - Tech stack from .memory_bank/tech_stack.md - Wor ```