Mycelium

The intelligent substrate beneath your code

</div>

A meta-workflow orchestration plugin for Claude Code that implements systematic development with compounding knowledge capture. Like a mycelium network that learns, adapts, and grows stronger with each connection, this workflow creates an intelligent knowledge layer that makes every development session more capable than the last.

Inspired by Conductor, Superpowers, and Compound Engineering principles.

Core Philosophy

Each unit of work should make subsequent units easier through compounding knowledge and systematic processes.

Like the mycelium network in nature that connects and nourishes an entire forest, this workflow creates an underground intelligence layer that:

🍄 Learns from patterns across sessions
🌐 Connects knowledge into a growing network
📈 Compounds capability with each use
🔄 Adapts to your development style

Workflow Overview

flowchart TD
    Start([Session Start]) --> CheckProject{Project<br/>Exists?}

    %% Phase -1: Bootstrap
    CheckProject -->|No| Bootstrap[-1: Project Bootstrap]
    Bootstrap --> Interactive[Interactive Setup<br/>One Question at a Time]
    Interactive --> CreateStructure[Create .mycelium/<br/>Directory Structure]
    CreateStructure --> GitInit[Initialize Git<br/>MANDATORY]
    GitInit --> Phase0

    %% Phase 0: Context Loading
    CheckProject -->|Yes| Phase0[0: Context Loading]
    Phase0 --> LoadContext[Load Project Context<br/>product.md, tech-stack.md]
    LoadContext --> LoadKnowledge[Load Institutional Knowledge<br/>solutions/, learned/]
    LoadKnowledge --> DiscoverCaps[Discover Capabilities<br/>All Plugin Scopes]
    DiscoverCaps --> Phase1

    %% Phase 1: Clarify (Capabilities already discovered in Phase 0)
    Phase1[1: Clarify Request] --> AskQuestions[Ask Clarifying Questions<br/>ONE at a time<br/>⚠️ Capabilities already cached from Phase 0]
    AskQuestions --> RequirementsClear{Requirements<br/>Clear?}
    RequirementsClear -->|No| AskQuestions
    RequirementsClear -->|Yes| Research{Need External<br/>Research?}
    Research -->|Yes| Phase15[1.5: Research]
    Research -->|No| Phase2
    Phase15 --> Phase2[2: Planning & Assignment]
    Phase2 --> LoadCachedCaps[Load Cached Capabilities<br/>from state.json<br/>discovered in Phase 0]
    LoadCachedCaps --> DecomposeRequest[Decompose User Request<br/>into Features]
    DecomposeRequest --> DecomposeFeatures[Decompose Each Feature<br/>into 2-5 min Tasks]
    DecomposeFeatures --> DefineGitStrategy[Define Git Strategy:<br/>• Feature = git branch<br/>• Multiple features = worktrees]
    DefineGitStrategy --> AssignCaps[Assign to Each Task:<br/>• Best-fit agent from cache<br/>• Relevant skills from cache<br/>• Model tier haiku/sonnet/opus]
    AssignCaps --> VerifyAssign[Verify Assignments Valid<br/>All exist in cached capabilities]
    VerifyAssign --> CheckFeatures{Multiple<br/>Features?}

    %% Phase 3: Implementation - Feature Level
    CheckFeatures -->|Yes| ParallelFeatures[3: Parallel Feature Development<br/>Git Worktrees]
    ParallelFeatures --> CreateWorktrees[Create Worktree Per Feature<br/>.worktrees/feature-name/<br/>branch: feature/name]
    CreateWorktrees --> SpawnFeatureAgents[Spawn Agent Per Feature<br/>Fresh Context Per Worktree]
    SpawnFeatureAgents --> FeatureTasks[Each Agent: Execute Tasks<br/>2-5 min units with TDD]
    FeatureTasks --> Phase45

    CheckFeatures -->|No| SingleFeature[3: Single Feature Development<br/>Git Branch Only]
    SingleFeature --> CreateBranch[Create Feature Branch<br/>git checkout -b feature/name]
    CreateBranch --> ExecuteTasks[Execute Tasks Sequentially<br/>2-5 min units with TDD]
    ExecuteTasks --> Phase45

    %% Phase 4.5: Verification
    Phase45[4.5: Verification] --> EvidenceCheck[Evidence-Based Check<br/>NO 'should work']
    EvidenceCheck --> RunTests[Run All Tests]
    RunTests --> Coverage[Check Coverage ≥80%]
    Coverage --> Lint[Run Linters]
    Lint --> Build[Verify Build]
    Build --> ContextCheck{Context<br/>>80%?}

    %% Phase 4.5B: Context Management
    ContextCheck -->|Yes| Phase45B[4.5B: Context Sync]
    Phase45B --> Summarize[Summarize Work<br/>≤500 tokens]
    Summarize --> UpdateProgress[Update progress.md]
    UpdateProgress --> SpawnFresh[Spawn Fresh Agent<br/>with Compressed Context]
    SpawnFresh --> Phase5
    ContextCheck -->|No| Phase5

    %% Phase 5: Two-Stage Review
    Phase5[5: Review] --> Stage1[Stage 1: Spec Compliance]
    Stage1 --> SpecCheck{Meets All<br/>Requirements?}
    SpecCheck -->|No| FixSpec[Fix Missing Requirements]
    FixSpec --> Phase4
    SpecCheck -->|Yes| Stage2[Stage 2: Code Quality Review<br/>Parallel Multi-Agent]
    Stage2 --> Security[Security Analysis<br/>OWASP + AI-specific]
    Stage2 --> Performance[Performance Analysis<br/>Algorithms + Bottlenecks]
    Stage2 --> Architecture[Architecture Review<br/>Patterns + Design]
    Security --> CombineReview[Combine Reviews]
    Performance --> CombineReview
    Architecture --> CombineReview
    CombineReview --> CriticalIssues{Critical<br/>Issues?}
    CriticalIssues -->|Yes| FixIssues[Fix Critical Issues]
    FixIssues --> Phase4
    CriticalIssues -->|No| Phase6

    %% Phase 6: Knowledge Capture
    Phase6[6: Finalization] --> Commit[Git Commit<br/>with Co-Author]
    Commit --> CreatePR[Create Pull Request<br/>MCP or CLI]
    CreatePR --> Phase6E[6E: Pattern Detection]
    Phase6E --> DetectPatterns[Detect 3+ Similar Solutions]
    DetectPatterns --> PromotePattern{Pattern<br/>Found?}
    PromotePattern -->|Yes| UpdateCritical[Update critical-patterns.md]
    UpdateCritical --> Phase6ESkill
    PromotePattern -->|No| Phase6F
    Phase6ESkill{Generate<br/>Skill?}
    Phase6ESkill -->|Yes| GenSkill[Auto-Generate Skill<br/>from Pattern]
    GenSkill --> Phase6F
    Phase6ESkill -->|No| Phase6F

    Phase6F[6F: Store Learned Knowledge] --> CaptureSolution[Capture Solution<br/>with YAML Validation]
    CaptureSolution --> CaptureDecision[Capture Decisions<br/>.mycelium/learned/decisions/]
    CaptureDecision --> CaptureConvention[Detect Conventions<br/>.mycelium/learned/conventions/]
    CaptureConvention --> UpdatePrefs[Update Preferences<br/>.mycelium/learned/preferences.yaml]
    UpdatePrefs --> TrackAntiPattern[Track Anti-Patterns<br/>.mycelium/learned/anti-patterns/]
    TrackAntiPattern --> RecordPrompt[Record Effective Prompts<br/>.mycelium/learned/effective-prompts/]
    RecordPrompt --> UpdateMetrics[Update Session Metrics]
    UpdateMetrics --> Done([Work Complete])

    %% Feedback Loops
    Done -.->|Knowledge<br/>Compounds| Phase0
    UpdateCritical -.->|Pattern<br/>→ Skill| Phase0
    RecordPrompt -.->|Learning<br/>→ Better Work| Phase4

    %% Styling
    classDef bootstrap fill:#e1f5ff,stroke:#01579b,stroke-width:2px
    classDef context fill:#fff9c4,stroke:#f57f17,stroke-width:2px
    classDef planning fill:#f3e5f5,stroke:#4a148c,stroke-width:2px
    classDef implementation fill:#e8f5e9,stroke:#1b5e20,stroke-width:2px
    classDef verification fill:#fff3e0,stroke:#e65100,stroke-width:2px
    classDef review fill:#fce4ec,stroke:#880e4f,stroke-width:2px
    classDef finalization fill:#e0f2f1,stroke:#004d40,stroke-width:2px
    classDef decision fill:#fff,stroke:#666,stroke-width:2px,stroke-dasharray: 5 5

    class Bootstrap,Interactive,CreateStructure,GitInit bootstrap
    class Phase0,LoadContext,LoadKnowledge,DiscoverCaps context
    class Phase1,Phase15,Phase2,LoadCachedCaps,DecomposeRequest,DecomposeFeatures,DefineGitStrategy,AssignCaps,VerifyAssign planning
    class ParallelFeatures,CreateWorktrees,SpawnFeatureAgents,FeatureTasks,SingleFeature,CreateBranch,ExecuteTasks implementation
    class Phase45,EvidenceCheck,RunTests,Coverage,Lint,Build,Phase45B,Summarize,UpdateProgress,SpawnFresh verification
    class Phase5,Stage1,Stage2,Security,Performance,Architecture,CombineReview,FixSpec,FixIssues review
    class Phase6,Commit,CreatePR,Phase6E,DetectPatterns,UpdateCritical,GenSkill,Phase6F,CaptureSolution,CaptureDecision,CaptureConvention,UpdatePrefs,TrackAntiPattern,RecordPrompt,UpdateMetrics finalization
    class CheckProject,Research,CheckFeatures,ContextCheck,SpecCheck,CriticalIssues,PromotePattern,Phase6ESkill decision

Key Features:

🔄 Feedback Loops: Knowledge compounds across sessions
⚡ Parallel Execution: Default mode with dependency management
🧪 Iron Law TDD: 11-step test-first cycle enforced
🎯 Evidence-Based: No "should work" - only verified proof
🧠 Context Management: Auto-sync at 80% threshold
📚 Knowledge Capture: Solutions → Patterns → Skills

Features

🎯 Comprehensive Workflow Phases

Phase -1: Project Bootstrap - Initialize projects with .mycelium/ structure
Phase 0: Context Loading - Load institutional knowledge + discover capabilities (cached)
Phase 1: Clarify Request - Ask clarifying questions, define scope
Phase 2: Planning & Assignment - Decompose request → features → 2-5 min tasks, assign capabilities
Phase 3: Feature Development - Parallel features via worktrees OR single feature via branch
Phase 4.5: Verification - Evidence-based testing (no "should work" claims)
Phase 5: Review - Two-stage review (spec compliance + code quality)
Phase 6: Finalization - Knowledge capture with feedback loops

🌳 Hierarchical Decomposition

Request → Features → Tasks

User Request: "Add user authentication"
  ↓
Features (user-facing capabilities):
  • Feature 1: OAuth Integration → branch: feature/oauth-integration
  • Feature 2: Login UI → branch: feature/login-ui
  • Feature 3: Session Management → branch: feature/session-mgmt
  ↓
Tasks (2-5 min atomic units):
  Feature 1 tasks: [1.1 Google OAuth, 1.2 GitHub OAuth, 1.3 Token storage]
  Feature 2 tasks: [2.1 Login form, 2.2 Logout button, 2.3 User profile]
  Feature 3 tasks: [3.1 Session storage, 3.2 Token refresh, 3.3 Logout]

Git Strategy:

One feature → git checkout -b feature/name
Multiple features → Git worktrees in .worktrees/ directory

Worktree Storage:

project-root/
├── .worktrees/              # Git worktrees (gitignored)
│   ├── feature-oauth/       # Feature 1 worktree + branch
│   ├── feature-login-ui/    # Feature 2 worktree + branch
│   └── feature-session/     # Feature 3 worktree + branch
├── .mycelium/               # Workflow metadata
├── .gitignore               # Includes .worktrees/
└── src/                     # Main source

Commands:

# Create worktree for each feature
git worktree add .worktrees/feature-oauth -b feature/oauth-integration
git worktree add .worktrees/feature-login-ui -b feature/login-ui
git worktree add .worktrees/feature-session -b feature/session-mgmt

# Each agent works in its own worktree
cd .worktrees/feature-oauth && [agent executes tasks]
cd .worktrees/feature-login-ui && [agent executes tasks]
cd .worktrees/feature-session && [agent executes tasks]

# Cleanup after merge
git worktree remove .worktrees/feature-oauth

📚 Knowledge Compounding System

Solutions Library (.mycelium/solutions/):

Pattern-based knowledge capture
Automatic promotion (3+ occurrences → critical-patterns.md)
Searchable by problem type and tags

Learning Store (.mycelium/learned/):

decisions/ - Architectural decisions with context
conventions/ - Detected code patterns
preferences.yaml - User preferences learned from corrections
anti-patterns/ - Mistakes to avoid
effective-prompts/ - Approaches that worked

🔄 Feedback Loops

Pattern → Skill: Auto-generate skills from recurring patterns
Work → Knowledge: Capture solutions, decisions, conventions
Context → Discovery: Auto-discover capabilities from all plugin scopes
Corrections → Learning: Learn preferences from user feedback

⚡ Parallel Execution (Default)

Git worktrees for isolated parallel tasks
Dependency management (blockedBy/blocks)
Task scheduler with DAG resolution
Model tier selection (haiku/sonnet/opus)

🧠 Context Window Management

Phase 4.5B: Automatic context sync
Progress files bridge context between sessions
Fresh agent spawning when context > 80%
Checkpoint-based resumption

Installation

From Local Directory

# Test locally
claude-code --plugin-dir /path/to/mycelium

# Or symlink to project plugins
ln -s /path/to/mycelium /path/to/project/.claude/plugins/mycelium

From Git Repository

# Clone to user plugins directory
git clone https://github.com/jason-hchsieh/mycelium ~/.claude/plugins/mycelium

Quick Start

1. Initialize a Project

/mycelium-setup

This creates the .mycelium/ directory structure:

.mycelium/
├── context/           # Project information
│   ├── product.md
│   ├── tech-stack.md
│   └── workflow.md
├── plans/             # Implementation plans
├── solutions/         # Documented solutions
│   └── patterns/
│       └── critical-patterns.md
├── learned/           # Learning store
│   ├── decisions/
│   ├── conventions/
│   ├── preferences.yaml
│   ├── anti-patterns/
│   └── effective-prompts/
├── state.json         # Session state
└── progress.md        # Human-readable progress

2. Create an Implementation Plan

/mycelium-plan "Add user authentication"

Creates a detailed plan with:

Task breakdown with complexity estimates
Dependency relationships (blockedBy/blocks)
Agent, skill, and model assignments
Acceptance criteria for each task
Test strategy

3. Execute the Plan

/mycelium-work

Executes tasks with:

Parallel worktree execution (default)
Iron Law TDD enforcement
Evidence-based verification
Automatic context sync

4. Review Implementation

/mycelium-review

Two-stage review:

Stage 1: Spec compliance (blocking gate)
Stage 2: Code quality (security, performance, architecture)

5. Capture Knowledge

/mycelium-capture

Captures:

Solutions with validated YAML frontmatter
Architectural decisions
Code conventions
Patterns (promotes to critical-patterns.md after 3+)
Generates skills from recurring patterns

Skill Reference

Primary Skills

Skill	Description	Interaction
`/mycelium-go [task] [--interactive]`	Full autonomous workflow (plan → work → review → capture)	Minimal (autonomous) or phase approvals (interactive)
`/mycelium-setup [--resume]`	Bootstrap project with mycelium structure	Interactive setup questions
`/mycelium-plan [description]`	Create implementation plan with TDD task breakdown	Clarifying questions
`/mycelium-work [task_id\|all]`	Execute tasks with strict TDD enforcement	Autonomous with progress updates
`/mycelium-review [--stage=1\|2\|all]`	Two-stage review (spec compliance + quality)	Report with decision point
`/mycelium-capture [track_id]`	Extract learnings and grow knowledge layer	Autonomous knowledge capture

Utility Skills

Skill	Description
`/mycelium-view [task\|track_id] [--verbose]`	Preview workflow plan without execution (dry-run mode with optional workflow diagram)
`/mycelium-status [--verbose]`	Display current progress and state
`/mycelium-continue [--full] [--track <id>]`	Resume interrupted work from checkpoint

Usage Patterns

🚀 Quick Start (Autonomous)

/mycelium-go "Add user authentication with JWT"
# Runs full workflow with minimal interaction

🎯 Controlled (Interactive)

/mycelium-go "Add user login" --interactive
# Asks for approval after each phase

🔧 Manual (Step-by-Step)

/mycelium-plan "Add user login"
/mycelium-work all
/mycelium-review
/mycelium-capture
# Full control over each phase

Architecture

All user-facing skills are thin wrappers (~20-60 lines each) that delegate to internal implementation skills. Other operations (worktrees, PR creation, context sync) are handled automatically by skills when needed.

Skills

All plugin functionality is implemented as skills (following the Agent Skills standard):

User-Facing Skills (slash commands)

Skill	Purpose
go	Full autonomous workflow (plan → work → review → capture)
plan	Create implementation plan with TDD task breakdown
work	Execute tasks with strict TDD enforcement
review	Two-stage review (spec compliance + quality)
capture	Extract learnings and grow knowledge layer
view	Preview workflow plan without execution (supports `--verbose` for diagram)
continue	Resume interrupted workflow from checkpoint (supports `--track` to switch plans)
status	Display current progress and state
setup	Bootstrap project with mycelium structure

Internal Skills (Claude-only, `user-invocable: false`)

Skill	Purpose	Used By
tdd	Iron Law TDD - RED → GREEN → REFACTOR enforcement	`mycelium-work`, `mycelium-go`
verification	Evidence-based validation (show actual test output)	`mycelium-work`, `mycelium-go`
context	Project context loading and management	All skills
recovery	Handle blockers and stuck states	As needed

Skill Architecture

User-facing skills contain full workflow logic directly:

mycelium-go contains the complete autonomous orchestration engine (plan → work → review → capture)
mycelium-plan, mycelium-setup, mycelium-capture contain their complete workflow logic (no wrapper pattern)
Internal skills (tdd, verification, context, recovery): Provide reusable domain expertise
Benefit: Each user-facing skill is self-contained and can be invoked independently

Agents

3 specialized agents for workflow tasks:

Agent	Purpose	Tools
spec-compliance-reviewer	Verify implementation matches requirements	Read, Grep, Glob, Bash
code-quality-reviewer	Assess security, performance, architecture	Read, Grep, Glob, Bash
learning-agent	Capture and analyze institutional knowledge	Read, Write, Grep, Glob, Bash

Hooks

SessionStart: Load project context + discover capabilities
Stop: Save session state

Configuration

Project-Level (`.mycelium/context/`)

product.md - Product vision and goals tech-stack.md - Technical stack details workflow.md - Development practices

Session State (`.mycelium/state.json`)

Tracks:

Current session ID and timestamps
Active tracks and worktrees
Discovered capabilities (agents, skills, MCPs)
Metrics (tasks completed/failed, context resets, interventions)

Project Maturity Modes

Configure in .mycelium/context/workflow.md:

Mode	TDD	Coverage	Review	Use Case
prototype	flexible	50%	minimal	Rapid experimentation
development	standard	80%	full	Active development
production	strict	90%	full	Production code
regulated	strict	95%	full + compliance	Regulated industries

Task Complexity Classification

Beyond simple T/S/M/L sizing:

complexity: M
integration_points: 2
domain_novelty: partially_known
test_complexity: integration
reversibility: requires_migration
uncertainty: medium

Dependency Management

Enable parallel execution:

Task 1.1: Setup database
  blockedBy: []
  blocks: [1.2, 1.3]

Task 1.2: User model
  blockedBy: [1.1]
  blocks: [2.1]

Plugin manages execution order automatically.

Model Tier Selection

# Per-task model assignment
Task 1.1: Database schema
  model: opus      # Complex architecture

Task 1.2: CRUD endpoints
  model: sonnet    # Standard implementation (default)

Task 1.3: Code review
  model: haiku     # Fast, cheap

External Plugin Integration

Works with these external plugins (optional):

Plugin	Provides	Fallback
github	github-mcp, GitHub skills	gh CLI
gitlab	gitlab-mcp, GitLab skills	glab CLI
gitea	gitea-mcp, Gitea skills	tea CLI
commit-commands	/commit, /commit-push-pr	git commands

Directory Structure

mycelium/
├── .claude-plugin/
│   └── plugin.json           # Plugin manifest
├── agents/
│   ├── review/
│   │   ├── spec-compliance.md
│   │   └── code-quality.md
│   └── learning-agent.md
├── skills/
│   ├── mycelium-go/          # User-facing: full autonomous workflow
│   ├── mycelium-plan/        # User-facing: create implementation plan
│   ├── mycelium-work/        # User-facing: execute tasks with TDD
│   ├── mycelium-review/      # User-facing: two-stage code review
│   ├── mycelium-capture/     # User-facing: extract learnings
│   ├── mycelium-continue/    # User-facing: resume interrupted workflow
│   ├── mycelium-status/      # User-facing: display progress
│   ├── mycelium-setup/       # User-facing: bootstrap project
│   ├── tdd/                  # Internal: TDD enforcement
│   ├── verification/         # Internal: evidence-based validation
│   ├── context/              # Internal: context window management
│   └── recovery/             # Internal: recovery protocols
├── hooks/
│   ├── hooks.json
│   └── scripts/
│       ├── session-start.sh
│       └── save-state.sh
├── templates/                # 13 template files
│   ├── project/              # Project setup templates (5 files)
│   │   ├── product.md.template
│   │   ├── tech-stack.md.template
│   │   ├── workflow.md.template
│   │   ├── CLAUDE.md.template
│   │   └── critical-patterns.md.template
│   ├── plans/
│   │   └── plan.md.template
│   ├── solutions/
│   │   └── solution.md.template
│   ├── state/
│   │   └── progress.md.template
│   └── gitignore/            # Language-specific .gitignore (4 files)
│       ├── node.gitignore
│       ├── python.gitignore
│       ├── go.gitignore
│       └── rust.gitignore
├── schemas/                  # 7 JSON schemas
│   ├── session-state.schema.json
│   ├── solution-frontmatter.schema.json
│   ├── plan-frontmatter.schema.json
│   ├── enums.json
│   └── ...
└── lib/                      # 13 utility modules
    ├── schema-validator.js   # Core infrastructure
    ├── state-manager.js
    ├── template-renderer.js
    ├── pattern-detector.js
    ├── discovery/            # Capability discovery (5 utilities)
    │   ├── capability-scanner.js
    │   ├── agent-discovery.js
    │   ├── skill-discovery.js
    │   ├── mcp-discovery.js
    │   └── cache-manager.js
    └── scheduler/            # Scheduling algorithms (4 utilities)
        ├── task-scheduler.js
        ├── dependency-graph.js
        ├── worktree-tracker.js    # State tracking only
        └── merge-analyzer.js       # Conflict analysis only

**Note:** The lib/ utilities follow "infrastructure-only" principle:
- **Removed**: Learning operations (now AI-powered via learning-agent)
- **Removed**: Agent spawning (now uses Task tool natively)
- **Kept**: Programmatic infrastructure (validation, discovery, scheduling algorithms)
- **Philosophy**: lib/ provides data structures/algorithms, not orchestration

Best Practices

1. Always Use TDD

Tests first, implementation second. No exceptions.

2. Document Solutions

Every non-trivial fix should be captured in .mycelium/solutions/.

3. Plan Before Implementing

Use /mycelium-plan to break down work systematically.

4. Leverage Parallelism

Design tasks with minimal dependencies for maximum parallelism.

5. Track Deviations

Update plans when requirements change, log reasons in deviations section.

6. Review Comprehensively

Use two-stage review: spec compliance first, code quality second.

7. Capture Knowledge

Run /mycelium-capture after completing work to build institutional knowledge.

Troubleshooting

"No .mycelium directory found"

Run /mycelium-setup to initialize the project.

"Cannot create worktree"

Ensure you're in a git repository and the branch doesn't already have a worktree.

"Tests not found"

Check test paths in your project configuration and ensure test framework is installed.

"Pattern detection not working"

Ensure solution files have proper YAML frontmatter with tags. Patterns detected at 3+ occurrences.

Development

Running Tests

# Unit tests
npm test

# Integration tests
npm run test:integration

# Coverage
npm run coverage

Contributing

Fork the repository
Create a feature branch
Follow the Iron Law TDD
Run /mycelium-review before submitting
Document solutions in PR description
Submit pull request

License

MIT

Credits

Inspired by:

Conductor - Workflow orchestration patterns
Superpowers - Knowledge compounding principles
Compound Engineering - Systematic improvement methodology

Built for Claude Code by Anthropic.

Support

Issues: GitHub Issues
Discussions: GitHub Discussions
Documentation: This README and inline documentation

Version

See plugin.json for current version. Uses semantic versioning.

Name	evidence-based-verification
Description	This skill should be used during Phase 4.5 (Verification) when the user requests "verify this", "check if it works", "run tests", or after implementation completion. Enforces evidence-based validation - no "should work" claims, requires actual test execution and proof of correctness.

evidence-based-verification

SKILL.md