Agent Handoff

Agent Handoff Documentation Generator

Overview

Autonomously generate comprehensive handoff documentation optimized for AI agent takeover. Analyzes your project and produces a complete Context Stack in docs/handoff/.

Core principle: Zero configuration. Fully autonomous. Machine-readable first.

Output: 8-document Context Stack in docs/handoff/ with master manifest.

The Process

Execute these 4 phases sequentially:

Phase 1: Discovery

Gather project information automatically:

1.1 Find Design Documents

Use Glob to discover existing documentation:

docs/**/*.md
README.md
ARCHITECTURE.md
*/README.md

Look for keywords: design, architecture, ADR, requirements, specification, RFC

1.2 Analyze Git History

git log --since="30 days ago" --pretty=format:"%h %s" --no-merges

Extract:

• Feature additions (feat:, add:, implement:)
• Major changes (refactor:, breaking:)
• Bug fixes (fix:, bug:)
• Architecture decisions from commit messages

1.3 Detect Project Type

File markers to check:

# Libraries/Packages
ls setup.py pyproject.toml Cargo.toml package.json

# Web Services
ls Dockerfile docker-compose.yml
find . -name "routes*" -o -name "api*" -o -name "handlers*"

# CLI Tools
grep -r "cobra" "click" "commander" --include="*.go" --include="*.py" --include="*.ts"

# Monorepos
find . -name "go.mod" -o -name "package.json" | wc -l
ls services/ packages/ apps/

Structure patterns:

• Go: cmd/, internal/, pkg/
• Library: src/, lib/, tests/
• Service: api/, handlers/, routes/

Classification logic:

IF (has Dockerfile + routes/) → web-service
ELSE IF (has setup.py + no HTTP) → library
ELSE IF (has main with arg parsing + no HTTP) → cli-tool
ELSE IF (multiple go.mod/package.json) → monorepo
ELSE → generic

Detection output format:

project_type: "web-service" # library | cli-tool | monorepo | generic
language: "go" # go | python | typescript | rust | etc
framework: "gin" # if detected (optional)
confidence: "high" # high | medium | low

1.4 Scan Codebase Structure

# Map directories (exclude hidden, node_modules, vendor)
find . -type d -not -path '*/\.*' -not -path '*/node_modules/*' -not -path '*/vendor/*'

# Identify entry points
find . -name "main.go" -o -name "main.py" -o -name "index.ts" -o -name "app.py"

# Count file types
find . -name "*.go" -o -name "*.py" -o -name "*.ts" -o -name "*.rs" | sort | uniq -c

Phase 2: Analysis

Extract structured information from discovered sources:

2.1 Extract Architecture Decisions

Sources:

• Existing ADR files in docs/adr/ or docs/architecture/
• Code comments containing: "IMPORTANT", "DO NOT", "WHY", "HACK"
• Git blame for major refactorings
• README sections about architecture

What to extract:

• Why this technology choice?
• Why this structure/pattern?
• What must NOT be changed and why?
• Known trade-offs and their rationale

2.2 Build Domain Dictionary

Extract terminology from:

1. Type Definitions

   # Go
   grep -r "type.*struct" --include="*.go"
   
   # Python
   grep -r "class " --include="*.py"
   
   # TypeScript
   grep -r "interface " --include="*.ts"
   

2. Database Schema

   # Find migrations
   find . -path "*/migrations/*" -o -path "*/schema/*"
   
   # Look for SQL files
   find . -name "*.sql"
   

3. API Schemas

   # OpenAPI/Swagger
   find . -name "openapi.yaml" -o -name "swagger.json"
   
   # GraphQL
   find . -name "schema.graphql" -o -name "*.gql"
   

Build mapping:

{
  "Term": {
    "code_synonyms": ["list", "of", "variations"],
    "database_column": "db_column_name",
    "type": "data_type",
    "api_field": "apiFieldName"
  }
}

2.3 Map System Boundaries

Only for distributed systems (web-service, monorepo):

1. Parse Docker Compose

   # Find docker-compose files
   find . -name "docker-compose*.yml"
   
   # Extract services, ports, dependencies
   

2. Find External API Calls

   # HTTP client usage
   grep -r "http.Client" "requests.get" "axios" "fetch"
   
   # Extract hostnames from code
   grep -r "https://" --include="*.go" --include="*.py" --include="*.ts"
   

3. Identify Databases

   # Database drivers
   grep -r "postgres" "mysql" "mongodb" "redis"
   
   # Connection strings (sanitized)
   grep -r "DATABASE_URL" "DB_HOST"
   

4. Generate Mermaid Graph

   graph TD
       A[Service Name :port] -->|protocol| B[Dependency]
       A -->|SQL| C[(Database)]
   

2.4 Discover Build Commands

Check these sources in order:

1. Makefile

   cat Makefile | grep "^[a-z].*:"
   

2. package.json scripts

   cat package.json | jq '.scripts'
   

3. CI/CD configs

   cat .github/workflows/*.yml
   cat .gitlab-ci.yml
   

4. Language defaults

   Go:       go build ./...
   Python:   pip install -e .
   Rust:     cargo build
   Node:     npm install
   

Extract:

• Build/compile command
• Lint command
• Test command
• Integration test command
• Deploy/run command

Phase 3: Generation

Generate all 8 documents using extracted data:

3.1 Generate Handoff_Manifest.yaml

Always generated first as master index.

project_name: "<extracted from git remote or dir name>"
generated_at: "<ISO 8601 timestamp>"
generator_version: "1.0.0"
project_type: "<from detection>"
language: "<from detection>"
framework: "<from detection or null>"
confidence: "<from detection>"

critical_context:
  - "./strategic-context/PRD_Machine_Readable.md"
  - "./strategic-context/Architecture_Decision_Records.md"
  - "./strategic-context/Domain_Dictionary.json"

execution_context:
  - "./operational-context/Agent_Runbook.md"
  <% if has_system_map %>
  - "./operational-context/System_Context_Map.mermaid"
  <% end %>
  - "./operational-context/Codebase_Walkthrough_Annotated.md"

guardrails:
  - "./guardrails/Test_Strategy_Matrix.md"

constraints:
  <% list extracted constraints from ADRs, comments %>

quality_gates:
  build_command: "<from discovery>"
  lint_command: "<from discovery or 'not configured'>"
  test_command: "<from discovery>"
  coverage_threshold: <from config or 80>

3.2 Generate PRD_Machine_Readable.md

Sources:

• Existing design docs
• README.md features section
• Git commits (features added)
• Code analysis (exported APIs, endpoints, commands)

Format:

# Product Requirements Document (Machine-Readable)

**Project:** <name>
**Type:** <project_type>
**Version:** <from package file or git tag>

## Purpose

<extract from README or design docs>

## Core User Flows

<Generate Gherkin-style scenarios from:>
- README examples
- Test descriptions
- Code comments
- API endpoint handlers

### Flow: <extracted flow name>

**Given:** <precondition>
**When:** <action>
**Then:** <expected result>

## Invariants

<Extract from:>
- Code comments with "MUST", "ALWAYS", "NEVER"
- Validation logic
- Test assertions
- Security checks

Examples:

- User passwords are never stored in plaintext
- All API endpoints require authentication except /health
- Database transactions use isolation level READ COMMITTED

## Negative Constraints

<Extract from:>
- Comments with "DO NOT"
- ADRs with "rejected alternatives"
- Code marked with "HACK" or "WORKAROUND"

Examples:

- Do not introduce new npm dependencies without approval
- Do not refactor the legacy auth system (external dependencies)
- Do not change database schema without migration

## Success Criteria

<Extract from:>
- Test coverage requirements
- Performance benchmarks
- README goals section

Examples:

- All endpoints respond in <200ms p95
- Test coverage >80%
- Zero critical security vulnerabilities

3.3 Generate Architecture_Decision_Records.md

Format:

# Architecture Decision Records

<For each discovered ADR or major decision:>

## ADR-<number>: <Title>

**Status:** <accepted | rejected | superseded>
**Date:** <from git or "unknown">

### Context

<Why was this decision needed?>

### Decision

<What was decided?>

### Consequences

**Positive:**

- <benefit 1>
- <benefit 2>

**Negative:**

- <trade-off 1>
- <trade-off 2>

### Do Not Refactor

<Extract code blocks that should NOT be changed:>

**File:** `path/to/file.go:line`
**Reason:** <why this ugly code exists>

Example:

- `internal/auth/session.go:45-67` - UUID v4 required for legacy Java service compatibility

3.4 Generate Domain_Dictionary.json

Format:

{
  "<Term>": {
    "code_synonyms": ["<variant1>", "<variant2>"],
    "database_column": "<column_name>",
    "type": "<data_type>",
    "api_field": "<apiFieldName>",
    "description": "<optional brief description>"
  }
}

Example:

{
  "User": {
    "code_synonyms": ["subscriber", "account_holder", "member"],
    "database_column": "usr_id",
    "type": "uuid",
    "api_field": "userId",
    "description": "Registered user account"
  },
  "Session": {
    "code_synonyms": ["auth_session", "login_session"],
    "database_column": "session_token",
    "type": "string",
    "api_field": "sessionId",
    "description": "Active authentication session"
  }
}

3.5 Generate System_Context_Map.mermaid

Only for: web-service, monorepo Skip for: library, cli-tool, generic

graph TD
    <% for each service %>
    <ID>[<Name> :<port>]
    <% end %>

    <% for each dependency %>
    <FROM> -->|<protocol>| <TO>
    <% end %>

    <% for each database %>
    <ID>[(<Database Type> :<port>)]
    <% end %>

    <% for each external API %>
    <ID>[<External API>]
    <% end %>

    style <main_service> fill:#4CAF50
    style <databases> fill:#FF9800
    style <external> fill:#2196F3

Example:

graph TD
    A[Web API :8080] -->|REST| B[Auth Service :50051]
    A -->|gRPC| C[Payment Service :50052]
    A -->|SQL| D[(PostgreSQL :5432)]
    B -->|Cache| E[(Redis :6379)]
    C -->|HTTP| F[Stripe API]

    style A fill:#4CAF50
    style D fill:#FF9800
    style E fill:#FF9800
    style F fill:#2196F3

3.6 Generate Agent_Runbook.md

Format:

# Agent Runbook

## Environment Setup

### Prerequisites

<Extract from README, Dockerfile, or detect:>

- <Language> <version>
- <Database> <version>
- <Other dependencies>

### Quick Start

```bash
# Clone repository
git clone <repo_url>
cd <project_name>

# Install dependencies
<from discovery: make setup, npm install, pip install, etc>

# Configure environment
<if .env.example exists>
cp .env.example .env.local
# Edit .env.local with your settings

# Run locally
<from discovery: make dev, npm start, go run, etc>

# Verify installation
<from discovery: make verify, npm test, etc>
```

Verification Commands

Build/Compile:

<from discovery or language default>

Lint:

<from discovery or "not configured">

Test:

<from discovery>

Integration Tests:

<from discovery or "not configured">

Coverage:

<from discovery or language-specific command>

Development Workflow

1. Create feature branch: git checkout -b feature/name
2. Make changes
3. Run tests: <test command>
4. Commit: git commit -m "feat: description"
5. Push: git push origin feature/name

Secret Management

<If .env.example exists:> Mock credentials available in .env.example for testing. Never commit real secrets to version control.

<If no .env.example:> Check documentation for required environment variables.

Common Issues

Service Dependencies

Start all services:

docker-compose up -d

#### 3.7 Generate Codebase_Walkthrough_Annotated.md

**Format:**

```markdown
# Codebase Walkthrough

## Project Structure

Directory Guide

/<directory>/

Purpose:

Rules:

Key Files:

• <file> -
• <file> -

Example patterns:

/cmd/

Purpose: Main application entry points Rules:

• One main.go per service
• No business logic here (only wiring)
• Keep it thin (<100 lines)

/internal/

Purpose: Private application code Rules:

• Not importable by external projects
• Contains business logic and domain models

/internal/core/

Purpose: Business logic and domain models Rules:

• Pure functions only
• No database calls directly
• No HTTP handling
• Testable in isolation

/internal/adapters/

Purpose: External integrations (DB, APIs, message queues) Rules:

• Retry logic mandatory
• Graceful degradation
• Circuit breakers for external APIs

/pkg/

Purpose: Public reusable utilities Rules:

• Framework-agnostic
• Thoroughly tested
• Well-documented

/api/

Purpose: API definitions (OpenAPI, protobuf) Rules:

• Version all breaking changes
• Keep backwards compatible

/tests/

Purpose: Integration and E2E tests Rules:

• Test real scenarios
• Use test doubles for external services

Entry Points

Main application: <path> Test suite: <path> Documentation: <path>

Important Files

Naming Conventions

3.8 Generate Test_Strategy_Matrix.md

Format:

# Test Strategy Matrix

## Test Types

| Test Type   | When Required         | Coverage Target | Location    | Example  |
| ----------- | --------------------- | --------------- | ----------- | -------- |
| Unit        | All new functions     | 90% branch      | `<pattern>` | `<file>` |
| Integration | New endpoints/modules | All happy paths | `<pattern>` | `<file>` |
| E2E         | Critical user flows   | Core journeys   | `<pattern>` | `<file>` |
| Performance | <if applicable>       | <target>        | `<pattern>` | `<file>` |

## Coverage Requirements

<Extract from:>
- CI config
- Test framework config
- Existing test patterns

**New features:** <X>% branch coverage minimum
**Bug fixes:** Must add regression test
**Refactoring:** No coverage decrease allowed

## Test Patterns

<Extract from existing tests:>

### Unit Tests

```<language>
<example from codebase>
```

Integration Tests

<example from codebase>

Mocking Strategy

Quality Gates

Before merging/deploying:

• ✅ All tests pass
• ✅ Coverage >= %
• ✅ No flaky tests
• ✅ Integration tests pass
• ✅ No test warnings

Running Tests

All tests:

<command>

Unit only:

<command>

Integration only:

<command>

With coverage:

<command>

Watch mode:

<command or "not supported">

CI/CD Integration

<Extract from .github/workflows or similar:>

Tests run on:

• Every push
• Every PR
• Before deploy

Pipeline:

#### 3.9 Generate generation-report.md

**Format:**

```markdown
# Handoff Documentation Generation Report

**Generated:** <timestamp>
**Generator Version:** 1.0.0
**Project:** <name>
**Project Type:** <type> (<language>)
**Detection Confidence:** <confidence>

## Documents Created

✅ Handoff_Manifest.yaml
✅ PRD_Machine_Readable.md (sources: <list>)
✅ Architecture_Decision_Records.md (found <N> ADRs)
✅ Domain_Dictionary.json (mapped <N> terms)
<% if has_system_map %>
✅ System_Context_Map.mermaid (<N> services, <N> databases, <N> external APIs)
<% else %>
⏭️  System_Context_Map.mermaid (skipped - not a distributed system)
<% end %>
✅ Agent_Runbook.md (found <N> commands)
✅ Codebase_Walkthrough_Annotated.md (analyzed <N> directories)
✅ Test_Strategy_Matrix.md (found <N> test types)

## Sources Analyzed

**Design Documents:** <N> files
<list paths>

**Git History:** <N> commits analyzed

**Code Files:** <N> files scanned

**Project Detection:**
- Type: <type>
- Language: <language>
- Framework: <framework or "none detected">
- Confidence: <confidence>

## Warnings

<if any issues occurred:>
⚠️  <warning message>

<if no issues:>
No warnings.

## Next Steps

1. **Review** `docs/handoff/Handoff_Manifest.yaml` for accuracy
2. **Validate** constraints in manifest match your requirements
3. **Check** `Domain_Dictionary.json` for terminology accuracy
4. **Test** commands in `Agent_Runbook.md` on fresh environment
5. **Commit** documentation to version control:
   ```bash
   git add docs/handoff
   git commit -m "Add agent handoff documentation"

Validation Checklist

Strategic Context

• PRD flows match actual user journeys
• Invariants are complete and accurate
• Negative constraints capture all "don't touch" areas
• ADRs explain key architectural decisions

Operational Context

<% if has_system_map %>

• System map shows all dependencies <% end %>
• Runbook commands work on fresh environment
• Codebase walkthrough explains all major directories
• Entry points are correctly identified

Guardrails

• Test strategy matches current practice
• Coverage requirements are enforced in CI
• Quality gates are accurate

Domain Dictionary

• All major entities are mapped
• Code/DB/API field mappings are correct
• No missing synonyms

Feedback

Found an issue? Suggestions for improvement? Regenerate with: /generate-handoff

### Phase 4: Output

Write all generated documents to disk and report results.

#### 4.1 Check for Existing docs/handoff/

```bash
if [ -d "docs/handoff" ]; then
    # Backup existing
    mv docs/handoff docs/handoff.backup.$(date +%Y%m%d-%H%M%S)
    echo "Backed up existing handoff docs"
fi

4.2 Create Directory Structure

mkdir -p docs/handoff/strategic-context
mkdir -p docs/handoff/operational-context
mkdir -p docs/handoff/guardrails

4.3 Write All Documents

Use Write tool for each document:

1. docs/handoff/Handoff_Manifest.yaml
2. docs/handoff/strategic-context/PRD_Machine_Readable.md
3. docs/handoff/strategic-context/Architecture_Decision_Records.md
4. docs/handoff/strategic-context/Domain_Dictionary.json
5. docs/handoff/operational-context/System_Context_Map.mermaid (if applicable)
6. docs/handoff/operational-context/Agent_Runbook.md
7. docs/handoff/operational-context/Codebase_Walkthrough_Annotated.md
8. docs/handoff/guardrails/Test_Strategy_Matrix.md
9. docs/handoff/generation-report.md

4.4 Present Summary

Print concise summary to user:

✅ Agent handoff documentation generated!

📁 Location: docs/handoff/

📄 Documents created:
  ✅ Handoff_Manifest.yaml (master index)
  ✅ Strategic Context (3 docs)
  ✅ Operational Context (3 docs)
  ✅ Guardrails (1 doc)
  ✅ Generation Report

🎯 Project: <name> (<type>, <language>)
🔍 Analyzed: <N> files, <N> commits, <N> design docs

📋 Next steps:
  1. Review docs/handoff/generation-report.md
  2. Validate docs/handoff/Handoff_Manifest.yaml
  3. Test commands in Agent_Runbook.md
  4. Commit with: git add docs/handoff && git commit

Important Guidelines

Autonomous Execution

• Never ask user questions - make informed decisions
• Use sensible defaults when information is missing
• Document assumptions in generation-report.md
• Warn about low-confidence detections

Machine-Readable First

• Use strict formats (YAML, JSON, Gherkin)
• Avoid ambiguous language
• Use assertions over descriptions
• Prefer lists and tables over prose

Completeness

• Generate all applicable documents
• Don't skip sections due to missing data
• Use "unknown" or "not detected" when appropriate
• Mark low-confidence extractions

Error Handling

• If project type detection fails → use "generic" mode
• If no design docs found → generate from code/README
• If no build commands found → use language defaults
• If existing docs/handoff/ → backup automatically

Quality

• Verify all file paths are correct
• Ensure manifest links point to real files
• Validate YAML/JSON syntax
• Test that Mermaid diagram is valid

Success Criteria

Must generate:

• ✅ All 8 documents (or 7 if System Map not applicable)
• ✅ Valid Handoff_Manifest.yaml with correct links
• ✅ generation-report.md with validation checklist

Must complete:

• ✅ In < 3 minutes for typical project
• ✅ Without user interaction
• ✅ With clear summary output

Must handle:

• ✅ Projects with no design docs
• ✅ Projects with existing docs/handoff/
• ✅ All 4 project types
• ✅ Multiple languages

Notes

• This skill runs in main conversation (not as subagent)
• Uses standard Claude Code tools (Glob, Grep, Read, Write, Bash)
• Output is git-friendly (all text files)
• Can be re-run safely (backs up existing docs)
• Optimized for AI agent consumption, not human reading