为 Claude Code 等 AI 编码智能体提供结构化的规范管理，在写代码之前生成可验证的 spec，防止 AI 幻觉。

https://github.com/ChenJazzyBoss/superSpec…

superSpec 让你在写代码前先生成一个结构化的 spec，然后用工具验证代码有没有按 spec 来。整个流程分七个阶段，工具会自动跑验证和归档，你只需要关注脑暴、写计划和实现。

ChenJazzyBoss/superSpec

Source: https://github.com/ChenJazzyBoss/superSpec

The problem

You tell Claude Code: “Add batch export to the system.”

It writes 500 lines of code. Tests pass. You merge.

Three days later you discover:

• PDF export was never mentioned but Claude assumed it
• Error handling covers 2 of 7 failure modes
• The “edge case” tests are actually happy-path tests with different data

The spec was in your head. Claude couldn’t see it.

The solution

superSpec sits between your intent and Claude’s code. It forces a structured spec to exist before any code is written, then validates that the code actually matches.

flowchart LR
    You["👤 You"] -- "batch export" --> BS["🧠 Brainstorm<br/>Route Evaluator"]
    BS -->|"🚀 Lightweight"| GS["📋 generate-spec"]
    BS -->|"📦 Full"| CH["📂 Change Dir"]
    GS --> V["✅ validate-spec<br/>+ SkillGuard"]
    CH --> V
    V -->|"verified spec"| CC["🤖 Claude Code"]
    CC -->|"code + evidence"| AR["📦 archive"]
    AR --> LS["📜 Living Specs"]

Quick start

# 1. Clone and build superspec
git clone https://github.com/ChenJazzyBoss/superSpec.git
cd superSpec
npm install
npm run build
npm run bundle-validate

# 2. Go to YOUR project and initialize
cd /path/to/your-project
node /path/to/superSpec/bin/superspec.js init

This creates .superspec/ and .claude/ in your project. Then in Claude Code:

/generate-spec

Claude will ask you questions, generate a structured spec, and validate it — before writing any code.

What it does

📋 Structured specs — Requirements use SHALL/MUST, scenarios use Given/When/Then. No ambiguity.

✅ Auto-validation — 9 built-in rules catch missing scenarios, vague words, and incomplete coverage.

🔍 Deep analysis — Optional --deep mode detects logical contradictions between scenarios and coverage gaps in requirements.

📊 Auto diagrams — Add <!-- DIAGRAM:flowchart --> to your spec, and Mermaid diagrams are generated and embedded automatically.

🔗 Source tracking — Add <!-- source: src/foo.ts --> to link specs to code. Get warned when source files change but spec doesn’t.

🏷️ Scenario classification — Scenarios are auto-tagged as normal/error/boundary. Missing error scenarios trigger warnings.

🔄 Delta changes — Describe what changed, not the whole file. Merge conflicts become impossible.

🛡️ Anti-hallucination — Red flag tables and checklists prevent Claude from skipping steps or faking completion.

🔍 SkillGuard — Programmatic detection of AI skip patterns. Run superspec guard to verify skill configuration.

📋 Init Template — Collect human context before spec generation. Solves the “AI can’t read your mind” problem.

📦 Multi-artifact validation — Validate module lists with superspec validate-modules. Circular dependency detection included.

🤖 Subagent orchestration — Dual-review pipeline: implement → spec-check → code-review. Every task.

CLI Commands

superspec init

Initialize superSpec in your project.

superspec init                  # Default setup
superspec init --interactive    # Interactive configuration
superspec init --ci             # Include GitHub Actions workflow
superspec init --template web-api   # Use Web API template
superspec init --list-templates     # List available templates

Creates .superspec/ directory structure, injects CLAUDE.md, copies templates and scripts.

superspec validate <spec>

Validate a spec file.

superspec validate batch-export              # By spec name
superspec validate .superspec/specs/batch-export/spec.md  # By file path
superspec validate batch-export --strict     # WARNING = failure
superspec validate batch-export --deep       # Logical consistency analysis

Output (JSON):

{
  "valid": true,
  "issues": [],
  "summary": { "errors": 0, "warnings": 0, "info": 0 },
  "scenarioTypes": {
    "requirements[0]": ["normal", "error", "boundary"]
  }
}

superspec ci

Batch validate all specs.

superspec ci              # Validate all specs
superspec ci --strict     # Strict mode
superspec ci --json       # JSON output

superspec generate <name>

Generate test code skeleton from spec.

superspec generate batch-export                          # TypeScript (default)
superspec generate batch-export -l python                # Python
superspec generate batch-export -o test/batch.test.ts    # Write to file

superspec update <name>

Incrementally update a spec via Delta JSON.

cat delta.json | superspec update batch-export
superspec update batch-export -f delta.json

superspec diff <name>

Compare current spec with historical version.

superspec diff batch-export                    # Compare with latest snapshot
superspec diff batch-export --from 2026-06-01  # Compare with specific version

superspec history <name>

List all historical snapshots of a spec.

superspec history batch-export

superspec archive <name>

Archive a completed change.

superspec archive add-pdf-export

superspec changes

List in-progress changes.

superspec changes

superspec guard

Check skill file’s anti-hallucination configuration.

superspec guard src/skills/generate-spec/SKILL.md    # Check skill config
superspec guard src/skills/validate-spec/SKILL.md --json  # JSON output

Verifies that skill files have proper red flag tables and HARD-GATE tags configured.

superspec validate-modules

Validate module list files.

superspec validate-modules modules.md                    # Validate module list
superspec validate-modules modules.md -p my-project      # With project name
superspec validate-modules modules.md --json             # JSON output

Checks module structure, detects circular dependencies, validates naming conventions.

superspec pipeline show

Display the default workflow definition.

superspec pipeline show

Output:

📋 superSpec 默认工作流

  阶段              类型     依赖
  ──────────────── ──────── ──────────────────
  brainstorm        可选     —
  generate-spec     必需     brainstorm
  validate-spec     必需     generate-spec
  write-plan        必需     validate-spec
  implement         必需     write-plan
  verify            必需     implement
  archive           必需     verify

superspec pipeline next <stage>

Query the recommended next step for a given stage.

superspec pipeline next brainstorm        # → generate-spec
superspec pipeline next validate-spec     # → write-plan
superspec pipeline next archive           # → "reached end of workflow"

superspec pipeline run <name>

Run the pipeline for a spec — auto-executes programmatic stages (validate-spec, archive) and outputs guidance for AI stages.

superspec pipeline run batch-export                     # Run from validate-spec
superspec pipeline run batch-export --from write-plan   # Resume from a specific stage

Execution records are persisted to .superspec/pipeline/<exec-id>.json.

superspec pipeline status [name]

View pipeline execution status.

superspec pipeline status               # Latest execution
superspec pipeline status batch-export  # Latest for this spec
superspec pipeline status --exec <id>   # By execution id

superspec pipeline list

List all pipeline execution records.

superspec pipeline list

superspec pipeline resume <exec-id>

Resume a failed pipeline execution from the failed stage.

superspec pipeline resume batch-export-20260611100000

superspec change

Manage change lifecycle (unified model for new features and modifications).

superspec change create <name>             # Create change directory with proposal
superspec change create <name> --why "..." # With description
superspec change list                      # List all changes
superspec change status <name>             # Show change phase and capabilities
superspec change apply <name>              # Merge delta specs into main specs
superspec change apply <name> --dry-run    # Validate without writing

superspec route

Evaluate user intent and recommend a path.

superspec route "add export button"          # → 🚀 Lightweight
superspec route "implement full auth" -c 3   # → 📦 Full
superspec route "modify export format"       # → 📦 Full (modification)
superspec route "export feature crashed"     # → 🔧 Debug

superspec uninstall

Remove all superSpec files from project.

superspec uninstall        # With confirmation
superspec uninstall -y     # Skip confirmation

Spec Format

A spec file uses structured Markdown:

# Feature Name

## Purpose

Description of what this feature does and why it's needed.
Must be at least 50 characters.

<!-- DIAGRAM:flowchart -->

## Requirements

### Requirement: Requirement Name
The system SHALL do something specific.

#### Scenario: Normal flow
Given some precondition
When some action
Then expected result

#### Scenario: Exception handling
Given some precondition
When an error occurs
Then error is handled gracefully

#### Scenario: Boundary condition
Given edge case data
When processing
Then system handles correctly

Spec Annotations

How it works

1. Generate a spec

/superspec:generate-spec

Claude asks about your requirements, then produces:

# Batch Export

## Purpose
System needs to export data in CSV, XLSX, and PDF formats
for various business workflows...

## Requirements

### Requirement: Format Support
System SHALL support CSV, XLSX, and PDF export formats.

#### Scenario: Normal flow - CSV export
Given user is on the data list page
When selecting CSV format and clicking export
Then system generates and downloads a CSV file

#### Scenario: Exception - Export failure
Given user is on the data list page
When an error occurs during export
Then system displays error message and logs the issue

2. Validate

# Basic validation (format + rules + scenario classification)
node .superspec/scripts/validate.js .superspec/specs/batch-export/spec.md

# Deep validation (+ logical consistency analysis)
node .superspec/scripts/validate.js .superspec/specs/batch-export/spec.md --deep

✅ valid: true
   errors: 0, warnings: 0, info: 0

   scenarioTypes:
     requirements[0]: [normal, error, boundary]

3. Implement

/superspec:write-plan
/superspec:subagent-dev

Claude creates a detailed plan, then implements each task with dual review.

4. Run the pipeline

superspec pipeline run batch-export

Automatically executes programmatic stages (validate-spec, archive) and guides you through AI stages (write-plan, implement, verify). Execution state is persisted — you can resume anytime.

5. Archive

/superspec:archive

Changes are recorded. Specs grow. History is preserved.

Features

Why superSpec?

The workflow

flowchart TD
    User["👤 User Input"] --> Router["🧭 Route Evaluator"]
    Router -->|"🚀 Lightweight<br/>(simple feature)"| Spec["📋 generate-spec"]
    Router -->|"📦 Full<br/>(complex / change)"| Change["📂 Change Dir<br/>proposal → delta-spec"]
    Router -->|"🔧 Debug<br/>(bug / failure)"| Debug["🔍 debug"]

    Change --> Spec
    Spec --> Validate["✅ validate-spec<br/>(auto)"]
    Validate -->|pass| Plan["📝 write-plan"]
    Validate -->|fail| Spec
    Plan --> Implement["🔨 implement"]
    Implement --> Verify["🧪 verify"]
    Verify -->|pass| Archive["📦 archive<br/>(auto)"]
    Verify -->|fail| Implement

    style Router fill:#f9f,stroke:#333
    style Validate fill:#9f9,stroke:#333
    style Archive fill:#9f9,stroke:#333
    style Debug fill:#ff9,stroke:#333

Three adaptive paths based on complexity assessment. Programmatic stages (validate-spec, archive) run automatically via pipeline run. AI stages output guidance and wait for completion.

Running the pipeline

Use superspec pipeline run to automate the workflow:

# Start pipeline (auto-runs validate-spec)
superspec pipeline run batch-export

# Output shows stage status + next step guidance
# 📋 管道执行: batch-export-20260611100000
#    ✅ validate-spec     completed   (150ms)
#    ⏳ write-plan        pending     — needs AI
#
# 📍 下一步操作指引:
#    阶段: write-plan
#    技能: write-plan
#    操作: 计划生成后，运行 superspec pipeline run <name> --from implement 继续管道

# After AI completes write-plan, resume pipeline
superspec pipeline run batch-export --from implement

# Check pipeline status anytime
superspec pipeline status batch-export

Programmatic stages (validate-spec, archive) run automatically. AI stages output guidance and wait for completion.

PipelineGuardRunner

The pipeline integrates SkillGuard at runtime. Every stage execution goes through anti-hallucination checks:

Stage starts → SkillGuard.beforeExecute() → check red flag table & HARD-GATE
Stage runs   → handler(context)
Stage output → SkillGuard.onOutput()      → detect skip patterns & red flags
Stage ends   → SkillGuard.onCompletion()  → verify evidence

If a skill file lacks a red flag table, the stage is blocked — not warned, blocked.

Anti-hallucination design

Every high-risk skill has:

Red flag table — Common excuses and why they’re wrong:

Completion checklist — Must check every box before declaring done:

• Verification command actually ran
• Full output read, exit code checked
• Zero failures

XML tag constraints — Behavioral guards in skill definitions:

<HARD-GATE>
No fresh evidence = no completion claim. No exceptions.
</HARD-GATE>

Project structure

superSpec/
├── bin/superspec.js          # CLI entry point
├── CLAUDE.md                 # Project AI behavior guide
├── src/
│   ├── cli/index.ts          # CLI commands
│   ├── core/
│   │   ├── validator.ts      # Validation engine
│   │   ├── spec-parser.ts    # Markdown parser
│   │   ├── spec-schema.ts    # Zod schemas
│   │   ├── module-schema.ts  # Module list schemas
│   │   ├── module-parser.ts  # Module list parser
│   │   ├── module-validator.ts # Module list validator
│   │   ├── deep-analysis.ts  # Logical consistency check
│   │   ├── diagram-generator.ts  # Auto Mermaid diagrams
│   │   ├── source-tracker.ts # Spec-code sync tracking
│   │   ├── delta-merge.ts    # Incremental spec updates
│   │   ├── change-lifecycle.ts   # Change directory management
│   │   ├── delta-spec-parser.ts  # Markdown delta spec parser
│   │   ├── specs-apply.ts    # Delta → main spec merge engine
│   │   ├── route-evaluator.ts    # Intent detection & path routing
│   │   ├── anti-rationalization/ # SkillGuard system
│   │   ├── pipeline/         # Skill orchestration pipeline
│   │   │   ├── executor.ts   # Pipeline execution engine
│   │   │   ├── runner.ts     # CLI runtime (run/status/list/resume)
│   │   │   ├── guard-runner.ts   # SkillGuard integration
│   │   │   ├── context.ts    # Stage-to-stage context passing
│   │   │   ├── conditions.ts # Pre/post condition checks
│   │   │   ├── retry.ts      # Failure classification & retry
│   │   │   └── workflow.ts   # Default 7-stage DAG definition
│   │   ├── rules/            # Validation rules
│   │   ├── diagrams/         # Diagram generators
│   │   └── config/           # Configuration system
│   ├── adapters/             # Test code generators
│   └── ci/                   # CI runner
├── templates/                # Project templates
│   ├── spec-template.md      # Spec scaffold
│   ├── init-spec-template.md # Init Template for context collection
│   ├── change/               # Change lifecycle templates
│   └── init-templates/       # Project-type templates
│       ├── general.md        # General project
│       ├── web-api.md        # Web API project
│       ├── cli.md            # CLI tool project
│       └── library.md        # Library/SDK project
├── test/                     # Test suite (558 tests)
└── dist/                     # Build output

Acknowledgments

superSpec stands on the shoulders of two excellent projects:

OpenSpec — The specs/changes/archive directory model and behavior contract spec format were directly inspired by OpenSpec’s approach to structured specification management. Their insight that “specs are living documents, not one-time artifacts” shaped superSpec’s core architecture.

superpowers-zh — The runtime behavioral constraints (XML tags, anti-hallucination patterns, subagent orchestration) were influenced by superpowers-zh’s methodology for controlling AI agent behavior during coding sessions.

Thank you both for open-sourcing your ideas. 🙏

Contributing

Found a bug? Open an issue.

Want to contribute? Fork, branch, PR. All contributions welcome.

Have ideas? Start a discussion.

License

MIT