StackMoss

Turn vague AI coding requests into a calibrated agent team that can actually ship.

StackMoss bootstraps runtime-native agent teams for Claude Code, Cursor, VS Code / Copilot, Codex, and Antigravity from one deterministic source of truth.

Most AI coding setups stop at "here are some rules."

StackMoss goes further: it generates a working team model with roles, governance, methodology, calibration flow, and eval scaffolding — so your agents can plan, implement, review, and verify work with much less drift.

🎯 What StackMoss Gives You

After bootstrap, your repo is no longer just "prompted". It gets a team that can:

• 🏗️ Turn rough product intent into a Tech Lead-first delivery flow
• 👥 Break work into 18 specialized role lanes — from TL, PM, BA through FE, BE, FS, MOBILE, DEVOPS, DATA, MLE, PE, and BRAND
• 📏 Enforce working discipline via 8 methodology modules: planning, TDD, debugging, evidence, review, Git workflow, execution loop, and code map maintenance
• 🎯 Smart trigger descriptions that activate the right role even when users don't use exact keywords
• 🔄 Recalibrate to the real repo instead of blindly following a template
• ✅ Run trigger quality evals and portable checks before you trust the team on real feature work

⚡ Before vs After

• Ask for a feature
• Agent guesses scope
• Implementation drifts
• Testing is inconsistent
• Docs and prompts go stale

• Intake captures intent + constraints
• Tech Lead calibrates to real repo
• Roles implement with word budgets
• QA verifies against acceptance criteria
• Eval loop catches behavioral drift

The Delivery Loop

flowchart LR
    A["🚀 Idea"] --> B["📋 Bootstrap team"]
    B --> C["🔍 TL calibrates to repo"]
    C --> D["📐 Feature breakdown"]
    D --> E["⚙️ Implement with methodology"]
    E --> F["✅ Check + eval"]
    F --> G["📦 Ship with confidence"]

🛡️ Built for Vibe Coding, But With Guardrails

StackMoss keeps the speed of vibe coding, but adds structure that prevents the usual chaos:

| What you get | Why it matters | |:---|:---| | Tech Lead-first workflow | One role owns architecture & coordination | | Replace-only config updates | Prevents accidental config bloat | | User confirmation gates | No surprise changes to shared team files | | Runtime-native output per platform | Not one-size-fits-all prompt blobs | | Trigger quality eval | Tests that skill triggers match real user prompts | | Word budgets per capability | Agents stay concise, not verbose |

👥 Agent Roles

StackMoss ships 18 roles organized into core team and specialized lanes:

| Role | ID | Capabilities | |:---|:---|:---| | Tech Lead | TL | Architecture, code review, context maintenance, planning | | Product Manager | PM | BRD discovery & brainstorming, roadmap, prioritization, stakeholder alignment | | Business Analyst | BA | Requirements elicitation, acceptance criteria | | Developer | DEV | Implementation, environment knowledge, debugging | | Frontend | FE | UI components, CSS/theming, accessibility, design quality & anti-slop | | Backend | BE | API endpoints, database schema, authentication | | Fullstack | FS | API-to-UI integration, scaffolding, performance | | Mobile | MOBILE | Native UI, bundle/memory/battery, sensors/permissions | | DevOps Engineer | DEVOPS | CI/CD, Docker/K8s/cloud, logging/alerting | | Data Engineer | DATA | ETL pipelines, data modeling, data quality | | ML Engineer | MLE | Model training, serving, monitoring, drift detection | | Prompt Engineer | PE | System prompts, eval harness, chain orchestration | | UI/UX Designer | UIUX | Design tokens, prototyping, usability reviews, design audit & atmosphere config | | Brand / Graphic | BRAND | Brand identity, asset generation, style guides | | Quality Assurance | QA | Test verification, regression checklists | | Documentation | DOCS | README updates, changelog | | Security-lite | SEC | Basic security checks | | DevOps-lite | OPS | Deploy and infra checks |

Roles are user-selectable via multiselect during intake (with smart defaults based on persona × project type). Pick 🤷 "Not sure" to skip — Tech Lead will shape the team after finalizing the product spec. Each role includes deep skill content (Iron Law, process, anti-patterns, checklist) that Tech Lead can enrich with project-specific patterns via ROLE_SKILL_OVERRIDES.md.

📖 Methodology Modules

| Module | Scope | What it enforces | |:---|:---|:---| | TDD Cycle | DEV, QA | Red → Green → Refactor discipline | | Debugging Protocol | DEV | Systematic error diagnosis | | Evidence Gate | ALL | Claims require proof | | Planning Protocol | TL | Parallel-friendly task clusters | | Review Reception | ALL | Accept feedback gracefully | | Git Workflow | ALL | Conventional commits, push before context loss | | Execution Loop | ALL | TL assigns → DEV builds → QA audits → Ship/Block |

🎯 Runtime Targets

One intake → five platforms. StackMoss compiles the same roles, capabilities, and methodology into the native skill format of each runtime:

| Target | Output | Notes | |:---|:---|:---| | Claude Code | CLAUDE.md + .claude/skills/<skill>/SKILL.md | Role-level skills with YAML frontmatter | | Cursor | .cursor/skills/<skill>/SKILL.md | Native Cursor Agent Skills | | VS Code / Copilot | .github/copilot-instructions.md | Copilot Instructions format | | Codex | AGENTS.md + .agents/skills/<skill>/SKILL.md | OpenAI Codex agent tree | | Antigravity | .agent/{skills,rules,workflows} | Rules + workflows + skills |

🚀 Quick Start

Install:

npm install -g stackmoss

Use in an existing repo:

cd /path/to/repo
stackmoss init

Or create a fresh workspace:

stackmoss new my-project
cd my-project

Then:

1. Answer the intake (choose your roles with Space, confirm with Enter)
2. Open your runtime and chat with Tech Lead to calibrate
3. Set up GitHub repo + MCP servers for your runtime (see README_AGENT_TEAM.md)
4. Run stackmoss check
5. Run stackmoss eval smoke

Full walkthrough: QUICK_START.md

📂 What Gets Generated

my-project/
├── team.md                  # Shared team config (CONSTITUTION + roles + facts)
├── FEATURES.md              # Feature breakdown with appetite sizing
├── NORTH_STAR.md            # Locked product direction
├── NON_GOALS.md             # Explicit scope boundaries
├── ROLE_SKILL_OVERRIDES.md  # Project-specific role calibration
├── CALIBRATE.md             # Calibration checklist for Tech Lead
├── README_AGENT_TEAM.md     # MCP + GitHub setup guide
├── stackmoss.config.json    # Machine-readable config
├── .claude/skills/          # Claude Code skills
├── .cursor/skills/          # Cursor skills
├── .agent/                  # Antigravity rules + workflows
├── .agents/skills/          # Codex skills
├── .github/                 # Copilot instructions
├── CODE_MAP.md              # Code structure + impact map (agent-maintained)
└── evals/                   # Rubric + test cases + trigger eval

⌨️ Command Reference

| Command | Description | |:---|:---| | stackmoss new <name> | Create a new StackMoss workspace | | stackmoss init [name] | Bootstrap StackMoss in the current repo | | stackmoss inject | Scan an existing repo and sync migration facts | | stackmoss resolve | Resolve migration questions | | stackmoss promote --confirm | Move from MIGRATING to OPERATIONAL | | stackmoss run <alias> | Run a command alias with patch proposal on failure | | stackmoss check | Validate config, budgets, and calibration readiness | | stackmoss eval [profile] [--grade] | Prepare or grade a live team evaluation | | stackmoss patch list/apply/reject | Manage patch proposals | | stackmoss upgrade | Merge CONSTITUTION only | | stackmoss map [--depth N] | Generate or refresh CODE_MAP.md |

🧑‍💻 Development

git clone https://github.com/max-rogue/Stackmoss.git
cd Stackmoss
npm install
npm test
npm run build

Current local verification:

• 321 passing tests across 42 test files
• TypeScript build passes
• Deterministic: no LLM calls in the test suite

🙏 Acknowledgments

• taste-skill by @lexnlin — design engineering principles (AI tells, bias correction, atmosphere dials) distilled into FE and UIUX roles

📜 License

MIT Copyright StackMoss