Skip to content
/

Conductor (MadAppGang)

claude-conductor-madappgang · MadAppGang/claude-code · ★ 280 · last commit 2026-03-15

CDD plugin inside a 14-plugin professional consulting marketplace; uniquely uses Claude Code's native Tasks API and git notes for plan tracking and audit trail.

Best whenTask synchronization between plan.md and the native Claude Code Tasks panel prevents status drift and provides UI-level visibility without reading files.
Skip ifMultiple tasks in-progress simultaneously (enforced: only one [~] at a time), Skipping TDD cycle phases
vs seeds
claude-conductor(Archetype 4, passive scaffold). MadAppGang conductor diverges by using Claude Code's native Tasks API for plan tracking…
Primitive shape 12 total
Commands 6 Skills 6
00

Summary

Conductor (MadAppGang) — Summary

The MadAppGang conductor plugin is one of 14 plugins in the MadAppGang Claude Code plugin marketplace (MadAppGang/claude-code), built and maintained by the Australian consultancy MadAppGang (creators of Claudish CLI and the 10xLabs platform). It implements the same Context-Driven Development methodology (Context → Spec & Plan → Implement) as the rbarcante/lackeyjb variants but as an embeddable module within a broader multi-plugin ecosystem. At v2.1.0, it ships 6 slash-commands (setup, new-track, implement, status, revert, help) with skills nested under each command rather than as a separate skills directory. The implementation commands use XML-style role/identity tags in their prompts and integrate with Claude Code's Tasks system for plan tracking. No hooks, no Python CLI, no Pattern Reference Layer — it is the leanest of the three CDD conductor variants. The broader marketplace it belongs to ships 14 plugins including a frontend plugin with 13 agents, a multimodel plugin with a team.json configuration, and MCP server integrations (Figma, Apidog, Chrome DevTools) — making the conductor plugin one piece of a much larger professional development toolkit aimed at Fortune 500 consulting teams.

01

Overview

Conductor (MadAppGang) — Overview

Origin

  • Author: MadAppGang (madappgang.com, 10xlabs.com.au)
  • Repo: https://github.com/MadAppGang/claude-code
  • Plugin path: plugins/conductor/
  • Version: 2.1.0 (plugin.json)
  • License: MIT
  • Stars: 280 | Forks: 32 | Contributors: unknown | Last commit: 2026-03-15
  • Description: "claude code plugins marketplace"

Philosophy

The README uses aggressive marketing language ("battle-tested AI workflows from the top 1% of developers," "forged in the fire of real deadlines"). The consultant framing is explicit:

"While most developers adopt new tools, we engineer competitive advantages." "Generated $2.3M in value through faster time-to-market on enterprise contracts" "Eliminated 156+ hours per sprint of repetitive development work"

The conductor plugin specifically implements the standard CDD philosophy: Context as a Managed Artifact — maintaining product.md, tech-stack.md, workflow.md in a conductor/ directory alongside code.

The larger marketplace context

MadAppGang ships 14 plugins total, including:

  • frontend (v2.8.0) — 13 agents, 6 commands, 3 skills, 3 MCP servers (Figma, Apidog, Chrome DevTools)
  • multimodel — multi-model team configuration via multimodel-team.json
  • code-analysis — semantic code search and analysis
  • orchestration — agent orchestration patterns
  • bun, go, seo, video-editing, instantly, nanobanana, statusline, agentdev, autopilot

The conductor plugin is the project management/planning layer within this larger ecosystem.

Conductor plugin philosophy

"Context as a Managed Artifact: Your project context (goals, tech stack, workflow) is documented and maintained alongside your code in the conductor/ directory."

Features: TDD workflow (Red/Green/Refactor), Phase Completion Protocol with verification reports, Git Notes audit trail, emergency procedures, and deployment workflow.

02

Architecture

Conductor (MadAppGang) — Architecture

Distribution

  • Type: Claude Code plugin within a multi-plugin marketplace
  • Install method:
    /plugin marketplace add MadAppGang/claude-code
    # then in project .claude/settings.json:
    { "enabledPlugins": { "conductor@mag-claude-plugins": true } }
    
  • Plugin update: /plugin marketplace update mag-claude-plugins
  • Required runtime: Claude Code

Plugin directory structure

MadAppGang/claude-code/
├── plugins/
│   ├── conductor/               # This plugin
│   │   ├── plugin.json          # v2.1.0 manifest
│   │   ├── README.md
│   │   ├── commands/
│   │   │   ├── setup.md
│   │   │   ├── new-track.md
│   │   │   ├── implement.md
│   │   │   ├── status.md
│   │   │   ├── revert.md
│   │   │   └── help.md
│   │   ├── skills/              # Per-command skill subdirs
│   │   │   ├── help/
│   │   │   ├── implement/
│   │   │   ├── new-track/
│   │   │   ├── revert/
│   │   │   ├── setup/
│   │   │   └── status/
│   │   └── templates/           # Workflow and context templates
│   ├── frontend/                # 13-agent frontend toolkit
│   ├── code-analysis/
│   ├── multimodel/
│   ├── orchestration/
│   └── ... (9 more plugins)
├── .claude/
│   ├── settings.json
│   ├── agents/
│   └── commands/
└── skills/                      # Global skills

Runtime directory (created at setup per project)

conductor/
├── product.md
├── product-guidelines.md
├── tech-stack.md
├── workflow.md
├── tracks.md
├── setup_state.json
├── code_styleguides/
└── tracks/<track_id>/
    ├── spec.md
    ├── plan.md
    └── metadata.json

Target AI tools

  • Primary: Claude Code
  • Marketplace model: multi-plugin; users enable individual plugins per project

Languages / runtimes

  • Markdown prompt files only
  • No Python, no Node.js required for the conductor plugin itself
03

Components

Conductor (MadAppGang) — Components

Commands (6)

Command Purpose
conductor-setup Initialize project context
conductor-new-track Create track with spec.md + plan.md
conductor-implement Execute tasks with TDD; phase completion verification; git notes audit trail
conductor-status Display progress
conductor-revert Git-aware undo
conductor-help Show available commands

Skills (per command, 6 subdirs)

Each command has a corresponding skill directory in skills/:

  • skills/setup/ — setup phase skills
  • skills/new-track/ — planning skills
  • skills/implement/ — implementation skills (TDD, verification, git notes)
  • skills/status/ — status reporting skills
  • skills/revert/ — revert operation skills
  • skills/help/ — help documentation skills

Sub-agents

No separately defined agent files (unlike rbarcante/lackeyjb). The implement command uses Claude Code's TaskCreate, TaskUpdate, TaskList, TaskGet tools to mirror plan.md tasks natively.

Hooks

None explicitly defined in the conductor plugin.

Scripts

None.

MCP servers (in conductor plugin)

None. (The broader MadAppGang marketplace's frontend plugin ships Figma, Apidog, Chrome DevTools MCP servers, but not the conductor plugin itself.)

Notable features per plugin.json tags

  • workflow, context, planning, task-tracking, git-aware, orchestration, project-management, development-workflow
  • Features: TDD workflow, Phase Completion Protocol, Git Notes audit trail, emergency procedures, deployment workflow

The broader marketplace (for context)

The full MadAppGang marketplace includes 14 plugins with a total of ~13 agents (frontend plugin), 3 MCP servers (frontend plugin), and a multi-model configuration system (multimodel-team.json).

05

Prompts

Conductor (MadAppGang) — Prompts

Excerpt 1: commands/implement.md — XML role pattern + Tasks API

---
name: conductor-implement
description: Execute tasks from track plan with TDD workflow and git commits
allowed-tools: AskUserQuestion, Bash, Read, Write, Edit, TaskCreate, TaskUpdate, TaskList, TaskGet, Glob, Grep
---

<role>
  <identity>Implementation Guide & Progress Tracker</identity>
  <expertise>
    - Task execution and status management
    - TDD workflow (Red/Green/Refactor)
    - Git commit integration with track references
    - Git Notes for audit trail
    - Workflow.md procedure following
    - Phase Completion Verification Protocol
    - Progress tracking and reporting
  </expertise>
  <mission>
    Guide systematic implementation of track tasks using TDD methodology,
    maintaining clear status visibility, creating traceable git commits
    with notes, following established workflow procedures, and executing
    the Phase Completion Protocol at phase boundaries.
  </mission>
</role>

<critical_constraints>
  <tasks_requirement>
    Use Tasks to mirror plan.md tasks.
    Keep Tasks and plan.md in sync.
    Mark tasks in BOTH when status changes.
  </tasks_requirement>

  <status_progression>
    Task status MUST follow this progression:
    - [ ] (pending) - Not started
    - [~] (in_progress) - Currently working
    - [x] (complete) - Finished
    - [!] (blocked) - Blocked by issue
    
    Only ONE task can be [~] at a time.
  </status_progression>

  <tdd_workflow>
    Red Phase: Create test file, write failing tests, confirm FAIL
    Green Phase: Write minimum code, confirm PASS
    Refactor Phase: Clean up while tests pass
  </tdd_workflow>
</critical_constraints>

Technique: XML-tagged role definition with <identity>, <expertise>, <mission> — a persona-declaration pattern. XML <critical_constraints> sections for hard rules. Integration with Claude Code native Tasks API (TaskCreate, TaskUpdate, TaskList, TaskGet) for progress tracking.


Excerpt 2: Plugin.json — structured metadata

{
  "name": "conductor",
  "version": "2.1.0",
  "description": "Context-Driven Development workflow for Claude Code. Manages project context (product.md, tech-stack.md, workflow.md) and guides structured development through: Context -> Spec & Plan -> Implement. Features TDD workflow (Red/Green/Refactor), Phase Completion Protocol with verification reports, Git Notes audit trail, comprehensive styleguides, emergency procedures, and deployment workflow.",
  "tags": ["workflow", "context", "planning", "task-tracking", "git-aware", "orchestration", "project-management", "development-workflow"],
  "commands": ["setup", "new-track", "implement", "status", "revert", "help"]
}

Technique: Comprehensive plugin description doubles as documentation and SEO for marketplace discovery.


Prompting techniques observed

  1. XML role declaration<role><identity>, <expertise>, <mission> structure agent persona
  2. XML constraint sections<critical_constraints> with nested typed rules
  3. Tasks API integration — explicit tool allowlist includes Claude Code's native task-tracking tools
  4. One-task-at-a-time rule — hard constraint that only ONE [~] task can exist at a time
  5. Companion skills pattern — each command has a matching skill directory for behavioral context
09

Uniqueness

Conductor (MadAppGang) — Uniqueness

differs_from_seeds

Closest seed is superbasicstudio/claude-conductor (Archetype 4, passive scaffold), but MadAppGang conductor is an active CDD plugin. It differs from rbarcante and lackeyjb variants primarily in two ways: (1) it uses Claude Code's native Tasks API (TaskCreate/TaskUpdate/TaskList/TaskGet) for plan tracking, where others use plain file reads; and (2) it ships as one of 14 plugins in a professional consulting marketplace, providing ecosystem context the standalone forks lack. The XML role declaration pattern (<role><identity><expertise><mission>) is distinct from rbarcante's XML-directive pattern and lackeyjb's plain markdown agents. Among seeds, the approach most resembles spec-driver (structured workflow with skills per command) but with CDD methodology rather than spec-first.

What makes MadAppGang conductor unique

  1. Native Tasks API integration — Claude Code tasks panel mirrors plan.md checkboxes; no other CDD variant does this
  2. Git Notes audit trail — track/phase/task metadata in git notes, not just commits
  3. Marketplace embedding — part of a 14-plugin professional ecosystem; can compose with frontend/orchestration/multimodel plugins
  4. XML role declarations<identity>/<expertise>/<mission> structure for agent prompts
  5. Help command/conductor-help as a first-class command; none of the other variants have this
  6. Command naming conventionconductor- prefix (dash) vs conductor: (colon) in other variants

Observable failure modes

  1. No hooks: unlike lackeyjb, no passive context loading; user must manually trigger all context reads
  2. No sub-agents: unlike rbarcante/lackeyjb, no specialist agent delegation; commands do everything inline
  3. Marketplace lock-in: the full ecosystem benefits require using MadAppGang's entire marketplace
  4. Marketing credibility gap: claims of "$2.3M value" and "156+ hours saved" are unverifiable; may mislead evaluators

Positioning vs competitors

MadAppGang conductor occupies the "enterprise consulting team" niche: teams adopting the full MadAppGang marketplace get CDD workflow management (conductor), frontend toolchain (frontend plugin with Figma+Chrome), and multi-model routing (multimodel plugin) as an integrated system. Individual developer teams are better served by rbarcante's standalone plugin.

04

Workflow

Conductor (MadAppGang) — Workflow

Phases

Phase Command Artifacts
1. Setup conductor-setup conductor/product.md, conductor/tech-stack.md, conductor/workflow.md, conductor/tracks.md, conductor/setup_state.json
2. Track Creation conductor-new-track conductor/tracks/<id>/spec.md, conductor/tracks/<id>/plan.md, conductor/tracks/<id>/metadata.json
3. Implementation conductor-implement Updated plan.md checkboxes, git commits with notes, phase verification reports
4. Status conductor-status (read-only)
5. Revert (optional) conductor-revert Git history modification

Approval gates

  1. Plan review: User reviews spec.md + plan.md before implementing
  2. Phase Completion Protocol: Manual verification step at each phase boundary with verification report
  3. AskUserQuestion: Interactive prompts at key decision points

TDD enforcement

Explicitly enforced via XML-role prompts in implement.md:

<tdd_workflow>
  Follow Test-Driven Development for each task:
  Red Phase: Create test, write tests, confirm FAIL
  Green Phase: Write minimum code, confirm PASS
  Refactor Phase: Clean up while tests pass
</tdd_workflow>

Task synchronization

Uses Claude Code's native Tasks API:

  • TaskCreate / TaskUpdate to mirror plan.md task status
  • Only ONE task can be [~] in_progress at a time
  • Status progression: [ ][~][x][!]

Git Notes audit trail

Commits linked to track/phase/task with git notes for traceability.

Emergency procedures

workflow.md and the implement skill include emergency procedures for blocked/failed tasks.

06

Memory Context

Conductor (MadAppGang) — Memory & Context

State storage

File-based in conductor/ directory, same pattern as rbarcante/lackeyjb:

File Contents
conductor/product.md Product vision and goals
conductor/product-guidelines.md Standards and conventions
conductor/tech-stack.md Technical preferences
conductor/workflow.md Development methodology (includes emergency procedures)
conductor/tracks.md Master track index
conductor/setup_state.json Setup resumption state
conductor/code_styleguides/ Language-specific style guides
conductor/tracks/<id>/spec.md Feature specification
conductor/tracks/<id>/plan.md Implementation plan with checkboxes
conductor/tracks/<id>/metadata.json Track state

Native task tracking

Uniquely among CDD variants: MadAppGang uses Claude Code's native Tasks API (TaskCreate, TaskUpdate, TaskList, TaskGet) to maintain a parallel task list synchronized with plan.md. This provides:

  • Native Claude Code task panel visibility
  • In-session task state without re-reading plan.md
  • Structured status transitions ([ ][~][x][!])

Git Notes audit trail

Commits include git notes with track/phase/task metadata — providing a structured audit trail in the git object store separate from commit messages.

Cross-session handoff

  • setup_state.json enables setup resumption
  • tracks.md provides cross-session work state

Memory persistence scope

  • Project-scoped: conductor/ within the repo

Context compaction

Not explicitly handled; relies on Claude Code's default context management.

07

Orchestration

Conductor (MadAppGang) — Orchestration

Multi-agent

No — no separately defined agent files. Commands are self-contained; no Task tool delegation to specialist agents (unlike lackeyjb's planner/implementer/reviewer pattern).

However, the conductor plugin exists within a larger marketplace where other plugins (particularly frontend with 13 agents and orchestration) provide multi-agent capabilities. The conductor plugin focuses on CDD workflow management, not agent orchestration.

Orchestration pattern

Sequential — commands execute their own logic directly. Tasks are tracked in plan.md and via native Tasks API. No sub-agent delegation.

Isolation mechanism

Git branches (inferred from workflow; git notes per commit).

Multi-model

No within the conductor plugin itself. The broader marketplace has a multimodel plugin with multimodel-team.json that can configure model assignments, but that is separate from the conductor plugin.

Execution mode

Interactive loop — user-triggered slash commands.

Prompt chaining

Yes — new-track produces spec.md + plan.md consumed by implement.

Task synchronization

The unique feature: Claude Code Tasks API keeps tasks synchronized between the prompt context and native task panel. Only one task [~] at a time — sequential execution enforced by constraint.

Crash recovery

Yes — setup_state.json for setup resumption. Task status preserved in both plan.md and native Tasks.

Cross-plugin composition (broader marketplace)

The marketplace's orchestration plugin provides multi-agent patterns that can compose with conductor's CDD workflow, enabling conductor for planning + orchestration for execution — but these are separate plugins, not an integrated system.

08

Ui Cli Surface

Conductor (MadAppGang) — UI & CLI Surface

Dedicated CLI

No standalone binary for the conductor plugin.

The broader MadAppGang ecosystem ships Claudish CLI (https://github.com/MadAppGang/claudish) as an external tool for running external AI models alongside Claude Code, but this is separate from the conductor plugin.

Local UI

None for the conductor plugin.

The broader marketplace integrates Chrome DevTools MCP server in the frontend plugin for browser-based validation — but this is the frontend plugin, not conductor.

IDE integration

  • Claude Code only
  • Marketplace installation: /plugin marketplace add MadAppGang/claude-code
  • Enable per-project in .claude/settings.json
  • Plugin update: /plugin marketplace update mag-claude-plugins
  • Slash commands: conductor-setup, conductor-new-track, conductor-implement, conductor-status, conductor-revert, conductor-help

Note: Command prefix is conductor- (with dash) vs conductor: (with colon) in rbarcante/lackeyjb variants.

Observability

  • Git Notes: audit trail in git object store
  • Phase verification reports: generated at phase boundaries
  • Native Tasks panel: task status visible in Claude Code UI
  • tracks.md: master progress index

Broader marketplace observability features

The statusline plugin provides status monitoring; code-analysis provides semantic code analysis — both can compose with conductor for enhanced observability.

Related frameworks

same archetype · same primary tool · same memory type

BMAD-METHOD ★ 48k

Provides a full agile delivery lifecycle with named expert-persona AI collaborators that elicit the human's best thinking rather…

Agent OS ★ 4.6k

Extracts implicit codebase conventions into token-efficient markdown standards files and injects them selectively into AI agent…

Claude Conductor ★ 367

Gives Claude Code a persistent, cross-linked, auto-analyzed documentation system so it retains codebase context across sessions.

Spec-Driver (Greenfield Spec-Driven Development) ★ 25

Prevents spec rot in AI-assisted development by making implementation changes flow back into evergreen, authoritative specs via…

Anthropic Knowledge Work Plugins ★ 16k

Role-specialized plugin bundles with live MCP connectors that turn Claude into a domain expert for enterprise knowledge workers.

Codex Integration for Claude Code (skill-codex) ★ 1.3k

Single Claude Code skill that handles Codex CLI invocation correctly (stdin blocking, thinking token suppression, session resume)…