All Projects

Multi-Agent Pipeline

8-phase AI development pipeline for Claude Code, Copilot CLI & Codex CLI. Issue → analysis → plan → TDD → rule-cited review → commit → PR in one command.

Drives a Jira issue or GitHub URL to a merged PR in one command (v16.18.0). 8 phases: Init · Analysis · Planning (approval gate) · TDD dev · CLI-aware parallel review (Fable + Sonnet on Claude Code, GPT + Opus + Sonnet on Copilot CLI, three effort-diverse GPT-5.6/5.4 reviewers on Codex CLI, then triage) · Test · Commit + PR · Report (PR body, Jira, Wiki, Confluence). Every loop is bounded and evidence-gated: deterministic gates before AI review, a 'passed' claim needs a log behind it, blocking findings can be confirmed by an auto-written repro test, and retry loops hard-stop at 3. Multi-repo orchestration with per-repo worktrees, per-phase cost ledger, cross-session learnings memory. Stack skills ship as versioned marketplace plugins. Runs natively on Claude Code, Copilot CLI and Codex CLI - macOS / Linux / Windows. Full version history in the repo CHANGELOG.

ArchitecturePrivate Repository
8
Phases
53
Commands
210
Skills
8
Sub-agents
Stable
Status

Tech Stack

Node.jsClaude CodeCopilot CLI210 Skills8 Sub-AgentsJSON Schemas

How It Works

Setup
🔑
Keychain
11 services · servicePatMap per identity
👤
Git Identity
identities[] · servicePatMap ownership
🔍
Repo Discovery
Local + Bitbucket + GitHub 90-day
🗺
Identity Routing
URL glob → identity auto-pick
📋
Stack Detect
iOS · Android · Backend · Frontend
🔒
Pre-commit
Secret scanning
Inputs (5 types)
GitHub Issue#316
GitHub URLgithub.com/.../issues/N
Jira IDPROJ-1234
Jira URLjira.../browse/...
Free Text"Fix dark mode bug"
Launch Modes (8 dispatch variants)
Claude/multi-agent "#316"Copilotmulti-agent "#316"
Normal0→1→2→3→4→5→6→7
Claude/multi-agent "PROJ-1234"Copilotmulti-agent "PROJ-1234"
Jira0→1→2→3→4→5→6→7
Claude/multi-agent:local "#316"Copilotmulti-agent-local "#316"
local0→1→2→3→4→6→7
Claude/multi-agent:autopilot "#316"Copilotmulti-agent-autopilot "#316"
autopilot0→1→2→3→4→6→7
Claude/multi-agent:local-autopilot "#316"Copilotmulti-agent-local-autopilot "#316"
autopilot + local0→1→2→3→4→6→7
ClaudeStep 7.5 → ShortCopilotStep 7.5 → Short
Short0→3→4→5→6→7
Claude/multi-agent:analysis "PROJ-2001"Copilotmulti-agent-analysis "PROJ-2001"
analysis0→1→2→4→6→7
ClaudeStep 1b → CorporateCopilotStep 1b → Corporate
analysis · corporate0→1→2→4→6→7
Helper Commands
setup
Keychain token + Git identity onboarding
jira
Browse your open Jira issues → pick → launch
issue
Browse unassigned GitHub issues → pick → launch
status
List tasks across all projects
log #N
Show task log
resume #N
Resume a stopped task
kill #N
Delete worktree, logs preserved
review
Review only the current diff
refactor
Autonomous refactoring with TDD safety net
channels
Multi-channel reporter — PR · Jira · Confluence · Wiki (multi-select, humanizer, v5.7+ replaces enrich)
test
UI Bug Hunter (simulator/emulator auto-detect)
stack [platform]
Swap platform skills on-the-fly
language [en|tr]
Toggle outputLanguage (assistant explanations). Picker prompts stay English.
scan
Skill security (4-tier · warn-only · CI strict)
search
Log full-text search (smart ranking · ripgrep)
sync
6-target sync (incl. plan-driven figma source pull)
update
Upgrade: git pull + install + migrate
clear-logs
Clean global log directory
purge
Full reset (double confirmation)
help
Usage guide and examples
UI Testing (Standalone)
🧪
test
Full simulator test - screenshot all screens
🌙
test "dark mode"
Dark mode bug test
♿
test "accessibility"
Accessibility audit - labels, tap targets, contrast
🔤
test "dynamic type"
Large text size test
📦
test "store-ready"
Full audit: visual + accessibility + archive compliance
🔐
test "biometric"
Face ID / Touch ID flow test
⚡
test "performance"
Launch time + scroll performance test
Stack Swap
Swap platform skills on-the-fly
stack iosstack androidstack backendstack mobilestack all
Skills (60+)
Pluggable knowledge modules injected into agents
🍎iOS · SwiftUI · UIKit · CoreData
🤖Android · Compose · Hilt · Room
⚙FastAPI · NestJS · Docker · CI/CD
🌐React · Next.js · Tailwind · Vue
✓Clean Code · TDD · Architecture · Security
🔧Git Actions · Figma · ASO · Audit Scripts
Pipeline Phases
0
Init
Multi-select picker
Identity routing
Fetch-fail 4-opt
N worktrees (serial)
Branch collision auto -v2
Step 0 language (en/tr)
1
Analysis
Parallel Explore
Fable model
Codebase scan
2
Planning
Task breakdown
Architecture
User approval
Todo repo tagging
3
Dev
TDD: Red→Green→Refactor
Per-repo worktree switch
Build verify
Max 3 retry
component → figma-to-component skill delegate
4
Review
Deterministic gates BEFORE AI review
Combined diff (repo headers)
crossRepo findings
Claude: Fable + Sonnet · Copilot: + GPT-5.4 → triage
5
Test
UI Bug Hunter
Dark/Light test
A11y audit
Simulator boot
6
Commit
Per-repo commit + push
Push-must-succeed (5 retries)
Cross-linked sibling PRs
Never auto-closes issues (4 approvals)
7
Report
Step 1: Channels dispatch — PR · Jira · Confluence · Wiki (multi-select, humanizer)
Autopilot ALWAYS pauses — 30-min timeout → session ends (resumable)
Step 2: agent-log · Quality & Metrics · telemetry
Step 3: Knowledge base + memory
Phase 0 - Interactive Init (8 Steps)
PARSE
Parse input, fetch from Jira/GitHub
→
PROJECT
Multi-select picker · recentGroups
→
BRANCH
TTL filter · collision auto -v2/-v3 · fetch-fail 4-opt
→
DESIGN?
Figma URL / screenshot (optional)
→
BRANCH NAME
Shared name · kebab rules · 50-char
→
IDENTITY
platformIdentityRouting · per-repo lookup
→
INSTRUCTION
Auto-detect .instructions/ directory
→
WORKSPACE
Multi-repo: N worktrees · serial · rollback
Sub-Agents (6)
🔍
Code Reviewer
Security, architecture, quality
🗺
Explorer
Codebase scan, patterns, impact areas
🍎
iOS Architect
Design decisions, patterns, modules
🤖
Android Architect
Design decisions, patterns, modules
⚙
Backend Architect
API design, data modeling, architecture
🛡
Security Auditor
Vulnerabilities and compliance
Component Pipeline (Figma → SwiftUI)
Figma URL detect→
Token extraction→
Config + View→
Modifiers→
Code Connect→
3-layer tests→
14-item review
Key Features
⚙ Phase 7 Report (Jira → Wiki+Figma → Confluence → Log → Knowledge)⚙ Skill Security Scanner (tiered · warn-only · CI strict · /multi-agent:scan)⚙ Log Search (smart ranking · filters · ripgrep · /multi-agent:search)⚙ Generic Figma-to-Component (iOS + Android)⚙ 4-Adapter Wiki (submodule / in-repo / github-wiki / separate-repo)⚙ Issue → Jira → Wiki Triad (autopilot-aware)⚙ Corporate Hosts in Setup (Jira / Confluence / Bitbucket inline prompt)⚙ Live Progress-Line Contract (quiet/normal/verbose)⚙ Plan-Driven Upstream Sync (--plan → review → apply)⚙ Overlay Protection (local-overlay: true frontmatter)⚙ Cross-CLI Parity (26 cmds + 37 figma skills identical)⚙ Build queue serialized with lock file⚙ Auto-detects iOS/Python/Node/Docker/Go⚙ Deterministic gates BEFORE AI review⚙ Multi-Repo Orchestration⚙ Platform Identity Routing⚙ Push-Must-Succeed (5 retries)⚙ Fetch-Fail 4-Option Flow⚙ Never auto-closes issues (4 approvals)⚙ PAT Audit Logging (SHA-256)⚙ 49 Smoke Suites (~870 assertions)⚙ Lazy file loading for token efficiency⚙ Pre-commit secret detection hook⚙ Autopilot Phase 2 Safety Classifier (v7.0.G · default-on · score≥50 → pause)⚙ Dynamic Trigger-Based Skill Loading (v7.0.I · 193-entry index · install.js --index-only)⚙ Per-Persona Model Routing (v6.1.D · preferredModel frontmatter · explorer → sonnet)⚙ OpenTelemetry Span Emission (v6.1.C · MULTI_AGENT_OTEL_SPANS=1 · JSONL)⚙ Phase 4 Disagreement Round (v6.1.B · opt-in · mixed-verdict rebuttal pass)⚙ Cost Summary in PR Body (v6.1.J · per-phase tokens + est. USD)⚙ Per-Repo File-System Memory Layer (v6.2.E · .multi-agent/memory/ · opt-in)⚙ Golden-Task Eval Harness (v6.2.F · 2 seed fixtures · wired to npm test)⚙ Skill Manifest Signing + Verify (v6.2.H · 193 SHA-256 · install-time warn-only)
Output
✓
Reviewed PR
Per-repo PRs · cross-linked siblings
📊
Report
Quality & Metrics · token rollup
🧠
Knowledge
Cross-session memory
📦
Components
Config + View + Tests
📁
Logs
~/.claude/logs/multi-agent/{project}/{task-id}/

Deep dive: architecture across three CLIs · install transforms, the 8 phases, every gate, and the shared state root

Live Replay - watch commands run end-to-end
Mode
Input
claude-code · /multi-agent PROJ-1234
▊
Tracker
8 phases · 2-model review · Report delivery
○0Init
○1Analysis
○2Planning
○3Dev
○4Review
○5Test
○6Commit
○7Report
running... · line 0/42
Simulated replay - actual pipeline emits the same progress-line shape (`→ verb object target`) at immediate flush.

Features

8-Phase Pipeline

Issue → analysis → plan (approval gate) → TDD dev → parallel review + triage → test → commit/PR → report. Every retry loop is bounded (hard-stop at 3).

CLI-Aware Parallel Review + Fable Triage

2 models on Claude Code (Fable + Sonnet), 3 on Copilot CLI (GPT + Opus + Sonnet); Fable triage filters false positives. Runs in a Short pipeline too, not just a Full one. Findings cite stable rule IDs from whichever rule registry declares scope over the changed files, so a review says SEC-01 instead of "I would have written it differently". Blocking findings can be confirmed by an auto-written repro test.

Deterministic Safety Gates

Build/lint/test/secret gates before AI review; a 'passed' claim needs a log behind it (default-FAIL evidence gate); diff risk scoring flags risky files and shrinking tests. Before the reviewers run, the applicable rule IDs are resolved and written to disk, so "was this applied completely" has a fixed denominator: an ID neither checked nor explicitly waived fails the stage.

Multi-Repo Orchestration

Per-repo worktrees, platform identity routing, integration builds, and file-based state so tasks resume across sessions and machines. A task worktree removes itself once its PR is open, salvaging its artefacts first and keeping the branch without checking it out, so your own working copy is never disturbed.

Cost & Telemetry

Per-phase token ledger, end-of-run cost breakdown naming the top cost driver, budget ceilings, and a model fallback contract (fable → opus → sonnet).

Cross-Session Memory

Persistent learnings ledger (rejected review preferences never get re-flagged), triage memory, and structured handoff blocks so long runs re-enter from durable artifacts, not conversation memory.

Available Commands

Claude Code (colon-form)Copilot CLI (dash-form)Description
Development
/multi-agent "task"multi-agent "task"

Full 8-phase pipeline with worktree + interactive confirmations.

/multi-agent:autopilot "task"multi-agent-autopilot "task"

Same pipeline, zero confirmations - fire and forget.

/multi-agent:jiramulti-agent-jira

Browse your open Jira issues in a terminal picker and launch.

/multi-agent:issuemulti-agent-issue

Browse unassigned GitHub issues, auto-assign, and launch.

Analysis
/multi-agent:analysis "feature"multi-agent-analysis "feature"

A document, not code: Init → Analysis → Synthesis → Review → Publish → Report.

Shared
/multi-agent:reviewmulti-agent-review

Standalone parallel review on a branch diff or PR, with triage.

/multi-agent:complaint-analysis [--file path]multi-agent-complaint-analysis [--file path]

Customer-complaint triage: log evidence per transaction id + read-only repo correlation, then a client/bff root cause with a fix plan and dev prompt, or a routing recommendation to the service owner.

/multi-agent:resume [#id]multi-agent-resume [#id]

Resume a paused/failed task from its last completed phase (file-based state).

/multi-agent:kill [#id]multi-agent-kill [#id]

Delete a task's worktree + branch; logs are preserved.

/multi-agent:prune-logsmulti-agent-prune-logs

Delete per-task logs (audit + metrics kept). Dry-run first.

/multi-agent:garbage-collectmulti-agent-garbage-collect

Sweep leftover /tmp scratch + worktree residue. Dry-run first.

/multi-agent:purgemulti-agent-purge

Full reset - every worktree, branch, state file and log. Double confirmation.

/multi-agent:setupmulti-agent-setup

First-run wizard - tokens, Git identity, per-project routing.

/multi-agent:helpmulti-agent-help

Bilingual usage guide - 47 live commands, the Full/Short depth question, preferences.

Back to overview