Batch 07 — Conductor Variants

Roster (10 + 1 seed reference)

What the name "Conductor" means in each project

Intra-batch patterns

Six of the ten entries share the "Context-Driven Development" (CDD) methodology — conductor/ directory with product.md + tech-stack.md + workflow.md + track-based spec+plan lifecycle. These are rbarcante, lackeyjb, MZWASHERE, MadAppGang, fcoury, and (loosely) the seed. Despite identical methodology names and similar artifact structures, they differ materially: rbarcante has a Python CLI + 9 commands + 6 agents + Pattern Layer; lackeyjb has 4 hooks + 3-agent delegation + no Python; MadAppGang uses Claude Code's native Tasks API; fcoury targets 3 AI tools simultaneously; and the seed just scaffolds templates. The four non-CDD entries (microsoft, conductor.build, code-conductor, codemachine-cli, claudiomiro) share no methodology — each is a distinct engineering paradigm (YAML runtime, desktop app, GitHub-swarm, JS-DSL, and autonomous DAG-loop respectively). The name "conductor" is clearly a meme pattern with no convergent meaning across the corpus.

Most interesting finds

1. microsoft/conductor — the only tool in the batch with deterministic Jinja2 routing (no LLM in orchestration), built-in FastAPI web dashboard, per-agent model assignment in YAML, and dual GitHub Copilot + Anthropic provider support. Its implement.yaml example shows a 6-agent multi-model workflow (Sonnet for coordination, Opus for deep implementation) that is the most sophisticated single workflow file in the entire corpus. The Microsoft institutional backing and SHA-256 install integrity check indicate production-grade readiness.

2. claudiomiro — the only tool with an explicit cross-run reflection + insight learning pipeline (ACE system). It stores curated lessons from completed tasks and injects them into future task contexts, creating a compounding learning loop. The combination of DAG parallel execution + local LLM cost optimization + multi-repo support positions it closer to Devin/SWE-agent than to Claude Code plugins. Nothing in the seed corpus does cross-run learning at this level of sophistication.

Items written as Tier C

• claude-conductor-mzwashere: spam fork of lackeyjb (exact byte-for-byte copy with broken download link README). 0 stars, 0 forks. Created same day as this analysis. Files written as stubs pointing to lackeyjb canonical. canonical: false.

Cross-references discovered

Direct lineage chain within batch

superbasicstudio/claude-conductor (seed)
  └─ rbarcante/claude-conductor (major feature expansion — adds Python CLI, Pattern Layer, ADRs, 9 commands, 6 agents)
       └─ lackeyjb/claude-conductor (simplification — removes Python CLI, adds 4 hooks, 3-agent delegation)
            └─ MZWASHERE/claude-conductor (spam fork — byte-for-byte copy of lackeyjb)
       └─ MadAppGang conductor plugin (independent rederivation — adds Tasks API, git notes, no hooks)
       └─ fcoury/conductor (cross-tool port — adds Gemini + Codex support, removes Python CLI)

Parallel lineage (same methodology, different inspiration)

• All CDD variants were also inspired by gemini-cli-extensions/conductor (not in this batch)
• lackeyjb's README explicitly says "Inspired by Conductor for Gemini CLI"
• fcoury also cites the Gemini CLI conductor extension

Functional equivalents across batch

• Conductor.build ↔ code-conductor: same worktree-based parallel agent model; Conductor.build is the commercial macOS UI; code-conductor is the open-source GitHub-native CLI
• microsoft/conductor ↔ codemachine-cli: both are workflow orchestration engines; microsoft uses YAML + Python SDK; codemachine uses JavaScript modules + subprocess spawning
• claudiomiro ↔ claude-flow (seed): both are continuous autonomous execution loops with memory persistence; claude-flow uses MCP + hive-mind consensus; claudiomiro uses subprocess + DAG + reflection

Dimensional comparison table