speccrew — Summary

Slug: speccrew GitHub: https://github.com/charlesmu99/speccrew Stars: 3 License: Apache-2.0 Version: v0.7.76 Language: JavaScript

One-Liner

npm-installable multi-agent spec factory — 8 specialized agents orchestrate PRD → feature design → API contract → system design → implementation → deployment → test with ISA-95 methodology and a 3-tier knowledge base.

Archetype

Skills-based multi-agent orchestrator with CLI dispatcher. Closest to trw-mcp in the current batch (multi-agent pipeline with human-confirmation gates) but ships as an npm package with no MCP layer and uses ISA-95 industrial modeling as its intellectual framework instead of value-oriented hooks.

Key Facts

• Install: npm install -g speccrew — speccrew CLI binary from bin/cli.js
• 8 agents: team-leader, product-manager, feature-designer, system-designer, system-developer, task-worker, test-manager, system-deployer
• 71 skills prefixed speccrew- covering planning, analysis, coding, testing, and knowledge management
• ISA-95 6-stage enterprise integration modeling as architectural methodology
• 3-tier knowledge base: L1 technology docs (knowledge/techs/), L2 business domain docs (knowledge/bizs/), L3 iteration artifacts (iterations/iXXX/)
• Iterations pattern: each requirement becomes an isolated project under iterations/iXXX-[name]/
• 4-stage knowledge pipeline: scan → parallel analysis → parallel summarize → aggregate
• Gates: all phase transitions require human confirmation
• Multi-language README: 10 languages (Chinese primary + EN, JA, RU, ES, DE, FR, PT-BR, AR, HI)
• Sub-agent auto-dispatch for context management; agents are dynamically created per tech stack

speccrew — Overview

What It Is

SpecCrew is an npm-installable multi-agent AI development orchestration system. Teams install it globally (npm install -g speccrew), initialize a project, then run the speccrew CLI to drive a complete software delivery pipeline from PRD authoring through deployment and testing.

The central thesis: treat software specification and delivery as an industrial process modeled on ISA-95 (the enterprise integration standard), where each "requirement" is an independent project iteration with its own isolated artifact directory.

Problem It Solves

Traditional AI-assisted development has agents that span too many concerns in a single context window, producing inconsistent output quality across phases. SpecCrew solves this by assigning hard-bounded specialists (product manager knows PRD, system designer knows API contracts, etc.) and enforcing phase gates with human confirmation before each handoff.

Target User

• Development teams wanting to go from idea → deployed software with AI doing the heavy lifting at each phase
• Teams with polyglot tech stacks (the dynamic agent creation per tech stack supports this)
• International teams: 10 README languages with Chinese as the primary, suggesting a strong Chinese developer community target

Distribution

• npm package: speccrew (globally installed)
• Repository: https://github.com/charlesmu99/speccrew
• Version: v0.7.76 (frequent minor releases)
• License: Apache-2.0

Positioning vs. Spec-Driven Seeds

• Unlike spec-kit (which outputs spec documents for human review), SpecCrew executes the specs through agents automatically
• Unlike openspec (OpenAPI-focused), SpecCrew generates API contracts as one phase of a broader pipeline
• Unlike spec-driver (which focuses on test-driven spec enforcement), SpecCrew treats testing as the last phase after deployment
• Closest analog in the seed set is a combination of taskmaster-ai (task breakdown) + bmad-method (role-based agent team) but with stronger industrial methodology grounding

speccrew — Architecture

High-Level Structure

speccrew/
├── bin/cli.js                    # CLI entry point
├── .speccrew/
│   ├── agents/                   # 8 agent definitions
│   │   ├── speccrew-team-leader.md
│   │   ├── speccrew-product-manager.md
│   │   ├── speccrew-feature-designer.md
│   │   ├── speccrew-system-designer.md
│   │   ├── speccrew-system-developer.md
│   │   ├── speccrew-task-worker.md
│   │   ├── speccrew-test-manager.md
│   │   └── speccrew-system-deployer.md
│   └── skills/                   # 71 skills (speccrew-* prefix)
├── knowledge/
│   ├── techs/                    # L1: Technology documentation
│   └── bizs/                    # L2: Business domain documentation
└── iterations/
    └── iXXX-[name]/             # L3: Per-requirement isolated artifacts

Knowledge Tier Architecture

3-tier knowledge base:

Knowledge Pipeline

4-stage pipeline for assembling context before each agent phase:

1. Scan — discover relevant L1/L2/L3 files
2. Parallel analysis — agents analyze files concurrently
3. Parallel summarize — distill key facts per file
4. Aggregate — combine into a unified context document

ISA-95 Methodology

SpecCrew uses ISA-95 (enterprise integration standard) as its modeling framework:

• 6-stage system modeling decomposition
• Clear separation between functional specifications and physical implementations
• Equipment hierarchy modeling applied to software component hierarchy

Agent Dispatch Model

• Team Leader coordinates phase transitions and dispatches to specialists
• Sub-agent auto-dispatch for context management: when context grows too large, the team leader spawns a sub-agent to handle the overflow
• Agents are dynamically created per tech stack — the system-designer agent generates stack-specific variants
• All phase transitions require human confirmation before proceeding

Execution Environment

• Runs inside Claude Code (.speccrew/agents/ and .speccrew/skills/ loaded by CC)
• No MCP server — agents communicate through filesystem artifacts
• iterations/iXXX-[name]/ acts as the state store per requirement

speccrew — Components

CLI Binary

Command: speccrew Entry point: bin/cli.js Distribution: npm global install

Agents (8 total)

Skills (71 total)

Skill naming convention: speccrew-{domain}-{action}

Domain prefixes:

• speccrew-pm-* — product management (requirement analysis, PRD structuring)
• speccrew-sd-* — system design (API contracts, data models)
• speccrew-dev-* — development (implementation, code review)
• speccrew-test-* — testing (test planning, test case generation)
• speccrew-knowledge-* — knowledge pipeline (scan, analyze, summarize, aggregate)
• Additional domain-specific skills for feature design and deployment

Knowledge Base

L1 — Technology docs (knowledge/techs/):

• Framework references, library documentation
• Tech stack decision records

L2 — Business domain (knowledge/bizs/):

• Domain glossary
• Business rules
• Stakeholder context

L3 — Iteration artifacts (iterations/iXXX-[name]/):

• PRD.md (product requirements)
• feature-design.md
• api-contract.yaml
• system-design.md
• implementation artifacts
• deployment config
• test results

Configuration Files

• speccrew.config.js / speccrew.json — project configuration
• .speccrew/agents/ — agent markdown definitions
• .speccrew/skills/ — skill markdown definitions

Output Artifacts Per Iteration

Each iterations/iXXX-[name]/ directory contains the complete lineage from PRD through test results for one requirement.

speccrew — Uniqueness

Differs from Seeds

SpecCrew is closest to a combination of bmad-method (role-based agent team with a product manager, architect, developer) and taskmaster-ai (task breakdown and tracking), but distinguished by:

1. ISA-95 industrial methodology — no other framework in the seeds or batch uses an enterprise integration standard as its intellectual backbone. This brings rigorous tier separation, level boundaries, and functional/physical distinction to AI-assisted development.

2. 3-tier knowledge base with 4-stage pipeline — the scan→parallel analysis→parallel summarize→aggregate pattern for context assembly is more sophisticated than any seed's context management. Unlike agent-os (flat markdown rules), SpecCrew's knowledge architecture separates technology context (L1), business context (L2), and iteration context (L3) with deliberate staleness management.

3. 71 skills with strict domain prefixing — speccrew-pm-*, speccrew-sd-*, speccrew-dev-*, speccrew-test-*, speccrew-knowledge-* — more skills than any seed, organized by agent role rather than by action type.

4. "Each requirement as independent project" pattern (iterations/iXXX/) — unlike taskmaster-ai's flat task list or bmad's single-project focus, SpecCrew treats every requirement as a completely isolated artifact tree. This enables parallel requirement development and clean audit trails.

5. 10-language README (Chinese-primary) — only multilingual framework in the seed set or this batch, indicating intentional non-English-first design.

Compared to spec-kit (outputs spec docs for human review): SpecCrew executes the specs through agents. Compared to openspec (API-focused): SpecCrew treats API contracts as one phase of a 7-phase pipeline. Compared to spec-driver (test-driven spec enforcement): SpecCrew tests last, not first.

Positioning

• Target user: Teams wanting full-pipeline automation from idea to tested deployment, with human gates at each phase
• Key differentiator: ISA-95 methodology + 3-tier knowledge architecture + per-requirement iteration isolation
• Honest gap: 3 GitHub stars and v0.7.76 suggest active development but pre-adoption; the industrial methodology may have a steep learning curve for typical web dev teams

Observable Failure Modes

1. 7 human gates — the "all transitions require confirmation" design maximizes safety but maximizes interruption; teams that want hands-off automation will be frustrated
2. No test-first — testing is phase 7, not phase 1; bugs introduced in implementation aren't caught until the end
3. Context overflow reliance — sub-agent auto-dispatch for context overflow is a runtime hedge, not a structured solution; behavior may be inconsistent
4. ISA-95 mismatch — industrial modeling concepts (equipment hierarchy, level boundaries) may confuse teams unfamiliar with the standard
5. CLI underdocumented — bin/cli.js exists but subcommand documentation is sparse