index

name	index
description	Build federated index from all child draft/ directories. Aggregates service-level context into root-level knowledge without deep code analysis. Use at monorepo root after services have been initialized with /draft:init.

Draft Index

You are building a federated knowledge index for a monorepo with multiple services.

Red Flags - STOP if you're:

• Running at a non-root directory in a monorepo
• Indexing services that haven't been initialized with /draft:init
• Overwriting root-level context without confirming with the user
• Aggregating without verifying each service's draft/ directory exists
• Skipping dependency mapping between services

Aggregate from initialized services only. Verify before overwriting.

---

Standard File Metadata

ALL generated files MUST include the standard YAML frontmatter. This enables refresh tracking, sync verification, and traceability.

Project Metadata File (single source of truth)

Before writing any file, update draft/metadata.json with current git state. This is the single source of truth for synced_to_commit and all git.* fields for every project-level artifact. Do NOT embed git fields in per-file frontmatter.

Use the --project-metadata flag (preferred):

DRAFT_TOOLS="${DRAFT_PLUGIN_ROOT:-$HOME/.claude/plugins/draft}/scripts/tools"
[ -d "$DRAFT_TOOLS" ] || DRAFT_TOOLS="$HOME/.cursor/plugins/local/draft/scripts/tools"
[ -d "$DRAFT_TOOLS" ] || DRAFT_TOOLS="$PWD/scripts/tools"
bash "$DRAFT_TOOLS/git-metadata.sh" --project-metadata \
    --project "$PROJECT" --generated-by "draft:index"

Per-File Metadata Template

Insert this stable YAML frontmatter at the top of every generated file (service-index.md, dependency-graph.md, tech-matrix.md, draft-index-bughunt-summary.md):

---
project: "{PROJECT_NAME}"
module: "root"
generated_by: "draft:index"
generated_at: "{ISO_TIMESTAMP}"
---

Note: generated_by uses draft:command format (not /draft:command) for cross-platform compatibility.

---

Pre-Check

ls draft/ 2>/dev/null

If draft/ does NOT exist at root:

• Announce: "Root draft/ directory not found. Run /draft:init at monorepo root first to create base context, then run /draft:index to aggregate service knowledge."
• Stop here.

If draft/ exists: Continue to lockfile check.

Lockfile Check

Before proceeding, check for a stale lock:

ls draft/.index-lock 2>/dev/null

• If draft/.index-lock exists and is less than 10 minutes old: Warn: "Previous indexing may be incomplete. Remove draft/.index-lock to proceed." Stop here.
• If draft/.index-lock exists and is older than 10 minutes: Remove it and continue.
• If no lock exists: Continue.

Create draft/.index-lock with the current timestamp before starting:

date -u +"%Y-%m-%dT%H:%M:%SZ" > draft/.index-lock

On completion (Step 9) or fatal error, remove the lock:

rm -f draft/.index-lock

Step 1: Parse Arguments

Check for optional arguments:

• --init-missing: Also initialize services that don't have draft/ directories
• bughunt [dir1 dir2 ...]: Run bug hunt across subdirectories with draft/ folders
  • If no directories specified: auto-discover all subdirectories with draft/
  • If directories specified: run bughunt only in those subdirectories (skip if no draft/)
  • Generate summary report at: draft/bughunt-summary.md

If bughunt argument detected: Skip to Step 1A (Bughunt Mode) instead of continuing to Step 2.

Step 1A: Bughunt Mode

This mode runs /draft:bughunt across multiple subdirectories and aggregates results.

1A.1: Determine Target Directories

If directories explicitly specified (e.g., bughunt dir1 dir2 dir3):

• Use provided directory list as targets
• Verify each directory exists
• Check each directory for draft/ subdirectory
• Warn and skip any directory without draft/

If no directories specified (e.g., just bughunt):

• Auto-discover all immediate child directories (depth=1)
• Filter for directories containing draft/ subdirectory
• Exclude patterns: node_modules/, vendor/, .git/, draft/, .*

# Example auto-discovery
for dir in */; do
  if [ -d "$dir/draft" ]; then
    echo "$dir"
  fi
done

Output:

Target directories for bughunt:
  - services/auth/
  - services/billing/
  - services/notifications/

1A.2: Execute Bughunt Per Directory

For each target directory:

1. Set working directory to <target-dir> for the bughunt scope. The AI agent should invoke /draft:bughunt with the target directory as the scope path, rather than using cd:

   /draft:bughunt
   → (scope prompt) → "Specific paths"
   → (paths prompt) → <target-dir>
   

2. Announce:

   Running bughunt in <target-dir>...
   

3. Let /draft:bughunt run its full workflow:

   • Report will be generated at <target-dir>/draft/bughunt-report-<timestamp>.md
   • Capture exit status (success/failure)

4. Record results:

   • Directory path
   • Total bugs found (by severity)
   • Report location
   • Any errors encountered

Note: Run bughunts sequentially, not in parallel, to avoid context conflicts.

1A.3: Parse Individual Reports

After all bughunts complete, read each generated report:

# For each target directory
cat <dir>/draft/bughunt-report-latest.md

Extract from each report:

• Branch and commit (from header)
• Summary table (bug counts by severity)
• Critical/High issue count
• Total issues count

1A.4: Generate Aggregate Summary Report

Create draft/bughunt-summary.md:

# Draft Index: Bughunt Summary

**Date:** YYYY-MM-DD HH:MM
**Mode:** [Auto-discovery | Explicit directories]
**Directories Scanned:** N

## Overview

| Directory | Critical | High | Medium | Low | Total | Report |
|-----------|----------|------|--------|-----|-------|--------|
| services/auth/ | 0 | 2 | 5 | 3 | 10 | [→](services/auth/draft/bughunt-report.md) |
| services/billing/ | 1 | 1 | 2 | 1 | 5 | [→](services/billing/draft/bughunt-report.md) |
| services/notifications/ | 0 | 0 | 1 | 2 | 3 | [→](services/notifications/draft/bughunt-report.md) |

**Grand Total:** X Critical, Y High, Z Medium, W Low

## Directories With Critical Issues

| Directory | Count | Details |
|-----------|-------|---------|
| services/billing/ | 1 | [→](services/billing/draft/bughunt-report.md#critical-issues) |

## Directories With No Issues

- services/api-gateway/
- services/user-service/

## Skipped Directories

| Directory | Reason |
|-----------|--------|
| services/legacy-tools/ | No draft/ directory found |
| services/experiments/ | No draft/ directory found |

## Next Steps

1. **Prioritize Critical Issues:** Review directories with Critical bugs first
2. **Review Individual Reports:** Click links above to see detailed findings
3. **Track Fixes:** Use `/draft:new-track` to create implementation tracks for fixes
4. **Re-run After Fixes:** Run `/draft:index bughunt` again to verify

---

*Generated by `/draft:index bughunt` command*

1A.5: Completion Report

---
              DRAFT INDEX BUGHUNT COMPLETE
---
Scanned: N directories
Completed: X successful
Skipped: Y (no draft/)
Failed: Z errors

Grand Total Bugs:
  Critical: W
  High: X
  Medium: Y
  Low: Z

Summary Report: draft/bughunt-summary.md

Directories requiring immediate attention:
  - services/billing/ (1 CRITICAL)
  - services/auth/ (2 HIGH)

---

STOP HERE if bughunt mode. Do not continue to Step 2 (normal indexing flow).

Step 2: Discover Services (Depth=1 Only)

Scan immediate child directories for service markers. Do NOT recurse beyond depth=1.

Service detection markers (any of these):

• package.json (Node.js)
• go.mod (Go)
• Cargo.toml (Rust)
• pom.xml or build.gradle (Java)
• pyproject.toml or requirements.txt (Python)
• Dockerfile (containerized service)
• src/ directory with code files

Exclude patterns:

• node_modules/
• vendor/
• .git/
• draft/ (the root draft directory itself)
• Any directory starting with .

# Example discovery (adapt to actual structure)
ls -d */ | head -50

Output: List of detected service directories.

Step 3: Categorize Services

For each detected service directory, check for draft/ subdirectory:

# For each service
ls <service>/draft/ 2>/dev/null

Categorize into:

• Initialized: Has draft/ with context files
• Uninitialized: No draft/ directory

Report:

Scanning immediate child directories...

Detected X service directories:
  ✓ Y initialized (draft/ found)
  ○ Z uninitialized

Initialized services:
  - services/auth/
  - services/billing/
  - ...

Uninitialized services:
  - services/legacy-reports/
  - services/admin-tools/
  - ...

Step 4: Handle Uninitialized Services

If init-missing argument is present:

1. For each uninitialized service, prompt:

   Initialize <service-name>/? [y/n/all/skip-rest]
   

2. If user selects:
   • y: Run /draft:init in that directory
   • n: Skip this service
   • all: Initialize all remaining without prompting
   • skip-rest: Skip all remaining uninitialized services

If init-missing argument is NOT present:

• Just report uninitialized services and continue
• Suggest: "Run /draft:index --init-missing to initialize these services"

Step 5: Aggregate Context from Initialized Services

For each initialized service, read and extract:

5.1 From <service>/draft/product.md:

• Service name
• First paragraph of Vision (summary)
• Target users (list)
• Core features (list)

5.2 From <service>/draft/.ai-context.md (or legacy <service>/draft/architecture.md):

• Key Takeaway paragraph (from ## System Overview)
• External dependencies (from ## External Dependencies)
• Exposed APIs or entry points (from ## Entry Points)
• Dependencies on other services (look for references to sibling service names)
• Critical invariants summary (from ## Critical Invariants, if available)

5.3 From <service>/draft/tech-stack.md:

• Primary language/framework
• Database
• Key dependencies

5.4 Create/Update <service>/draft/manifest.json:

{
  "name": "<service-name>",
  "type": "service",
  "summary": "<first line of product vision>",
  "primaryTech": "<main language/framework>",
  "dependencies": ["<other-service-names>", "<external-deps>"],
  "dependents": [],
  "team": "<if found in docs>",
  "initialized": "<date>",
  "lastIndexed": "<current-date>"
}

Step 6: Detect Inter-Service Dependencies

Analyze extracted data to build dependency graph:

1. Look for service name references in each service's architecture.md
2. Look for API client imports or service URLs in tech-stack.md
3. Look for mentions in product.md that reference other services
4. Graph-enriched detection (if individual services have draft/graph/ directories):
   • Read each service's draft/graph/architecture.json .routes to map which service defines vs consumes which endpoints
   • Cross-reference proto consumers with proto producers to build precise inter-service dependency edges
   • Read draft/graph/architecture.json (.packages) per service for internal module structure
   • This provides deterministic, code-level dependency data that supplements the heuristic name-matching above

Build a dependency map:

auth-service: [] # no dependencies on other services
billing-service: [auth-service]
api-gateway: [auth-service, billing-service]

Step 6.1b: Cycle Detection

Cycle detection: Walk the dependency graph depth-first from each service. If a cycle is detected (service A depends on B depends on ... depends on A), emit a > WARNING: Circular dependency detected: A → B → ... → A line in dependency-graph.md, mark the back-edge with circular: true in manifest.json's dependency entry, and continue processing. Do not fail on cycles — report and proceed.

Step 6.2: Resolve Dependents (Reverse Lookup)

For each service S, iterate all other services' dependencies arrays. If S appears in another service's dependencies, add that service to S's dependents array. Write the updated manifest.json for each service.

Step 7: Generate Root Aggregated Files

7.1 Generate draft/service-index.md

Use the following inline template:

# Service Index

> Auto-generated by `/draft:index` on <date>. Do not edit directly.
> Re-run `/draft:index` to update.

## Overview

| Metric | Count |
|--------|-------|
| Total Services Detected | X |
| Initialized | Y |
| Uninitialized | Z |

## Service Registry

| Service | Status | Tech Stack | Dependencies | Team | Details |
|---------|--------|------------|--------------|------|---------|
| auth | ✓ | Go, Postgres | - | @auth-team | [→](../services/auth/draft/.ai-context.md) |
| billing | ✓ | Node, Stripe | auth | @billing | [→](../services/billing/draft/.ai-context.md) |
| legacy-reports | ○ | - | - | - | Not initialized |

## Uninitialized Services

The following services have not been initialized with `/draft:init`:
- `services/legacy-reports/`
- `services/admin-tools/`

Run `/draft:index --init-missing` or initialize individually with:
```bash
cd services/legacy-reports && /draft:init

### 7.2 Generate `draft/dependency-graph.md`

```markdown
# Service Dependency Graph

> Auto-generated by `/draft:index` on <date>. Do not edit directly.

## System Topology

```mermaid
graph LR
    subgraph "Core Services"
        auth[auth-service]
        billing[billing-service]
        users[user-service]
    end

    subgraph "Edge"
        gateway[api-gateway]
    end

    subgraph "Background"
        notifications[notification-service]
        reports[report-service]
    end

    gateway --> auth
    gateway --> billing
    gateway --> users
    billing --> auth
    notifications --> users
    reports --> billing

Dependency Matrix

Service	Depends On	Depended By
auth-service	-	billing, gateway, users
billing-service	auth	gateway, reports
user-service	auth	gateway, notifications
api-gateway	auth, billing, users	-

Dependency Order (Topological)

1. auth-service (foundational - no internal dependencies)
2. user-service (depends on: auth)
3. billing-service (depends on: auth)
4. notification-service (depends on: users)
5. report-service (depends on: billing)
6. api-gateway (depends on: auth, billing, users)

This ordering helps when planning cross-service changes or understanding impact.

### 7.3 Generate `draft/tech-matrix.md`

```markdown
# Technology Matrix

> Auto-generated by `/draft:index` on <date>. Do not edit directly.

## Common Stack (Org Standards)

Technologies used by majority of services:

| Technology | Usage | Services |
|------------|-------|----------|
| PostgreSQL | Database | auth, billing, users (85%) |
| Redis | Caching | auth, gateway, notifications (60%) |
| Docker | Containerization | all (100%) |
| GitHub Actions | CI/CD | all (100%) |

## Technology Distribution

### Languages

| Language | Services | Percentage |
|----------|----------|------------|
| Go | auth, users, gateway | 45% |
| TypeScript | billing, notifications, reports | 45% |
| Python | ml-service, analytics | 10% |

### Databases

| Database | Services |
|----------|----------|
| PostgreSQL | auth, billing, users, reports |
| MongoDB | notifications, analytics |
| Redis | auth, gateway (cache only) |

## Variance Report

Services deviating from org standards:

| Service | Deviation | Reason |
|---------|-----------|--------|
| ml-service | Python instead of Go/TS | ML ecosystem |
| analytics | MongoDB instead of Postgres | Time-series workload |

Placeholder Detection

A file is considered a placeholder if it contains the marker <!-- AUTO-GENERATED --> or is smaller than 100 bytes. Placeholders may be overwritten without confirmation. Non-placeholder files require user confirmation before overwriting.

7.4 Synthesize draft/product.md (if not exists or is placeholder)

Read all service product.md files and synthesize:

# Product: [Org/Product Name]

> Synthesized from X service contexts by `/draft:index` on <date>.
> Edit this file to refine the overall product vision.

## Vision

[Synthesized from common themes across service visions - one paragraph describing what the overall product/platform does]

## Target Users

<!-- Aggregated from all services, deduplicated -->
- **End Users**: [common user types across services]
- **Developers**: [if developer-facing APIs exist]
- **Operators**: [if ops/admin services exist]

## Service Capabilities

| Capability | Provided By | Description |
|------------|-------------|-------------|
| Authentication | auth-service | User identity, JWT, OAuth |
| Payments | billing-service | Stripe integration, invoicing |
| API Access | api-gateway | Rate limiting, routing |

## Cross-Cutting Concerns

<!-- Extracted from common patterns across services -->
- **Authentication**: All services validate via auth-service
- **Observability**: [common logging/tracing approach]
- **Data Privacy**: [common compliance patterns]

7.5 Synthesize draft/architecture.md (if not exists or is placeholder)

# Architecture: [Org/Product Name]

> Synthesized from X service contexts by `/draft:index` on <date>.
> This is a system-of-systems view. For service internals, see individual service contexts.

## System Overview

**Key Takeaway:** [One paragraph synthesizing overall system purpose from service summaries]

### System Topology

```mermaid
graph TD
    subgraph "External"
        Users[Users/Clients]
        ThirdParty[Third-Party Services]
    end

    subgraph "Edge Layer"
        Gateway[API Gateway]
        CDN[CDN/Static]
    end

    subgraph "Core Services"
        Auth[Auth Service]
        Billing[Billing Service]
        Users2[User Service]
    end

    subgraph "Background"
        Notifications[Notifications]
        Reports[Reports]
    end

    subgraph "Data Layer"
        Postgres[(PostgreSQL)]
        Redis[(Redis)]
        Queue[Message Queue]
    end

    Users --> Gateway
    Gateway --> Auth
    Gateway --> Billing
    Gateway --> Users2
    Billing --> ThirdParty
    Auth --> Postgres
    Billing --> Postgres
    Notifications --> Queue
    Reports --> Queue

Service Directory

Service	Responsibility	Tech	Status	Details
auth-service	Identity & access management	Go, Postgres	✓ Active	→ context
billing-service	Payments & invoicing	Node, Stripe	✓ Active	→ context

Shared Infrastructure

Component	Purpose	Used By
PostgreSQL	Primary datastore	auth, billing, users
Redis	Caching, sessions	auth, gateway
RabbitMQ	Async messaging	notifications, reports
Stripe	Payment processing	billing

Cross-Service Patterns

Pattern	Description	Services
JWT Auth	All services validate JWT via auth-service	all
Event-Driven	Async events via message queue	notifications, reports

Notes

• For detailed service architecture, navigate to individual service contexts
• This file is regenerable via /draft:index
• Manual edits to non-synthesized sections will be preserved on re-index

### 7.6 Synthesize `draft/tech-stack.md` (if not exists or is placeholder)

```markdown
# Tech Stack: [Org/Product Name]

> Synthesized from X service contexts by `/draft:index` on <date>.
> This defines org-wide standards. Service-specific additions are in their local tech-stack.md.

## Org Standards

### Languages
- **Primary**: [most common language] — [X% of services]
- **Secondary**: [second most common] — [Y% of services]

### Frameworks
- **API**: [common API framework]
- **Testing**: [common test framework]

### Infrastructure
- **Database**: PostgreSQL (standard), MongoDB (approved for specific use cases)
- **Caching**: Redis
- **Messaging**: RabbitMQ / SQS
- **Container**: Docker
- **Orchestration**: Kubernetes

### CI/CD
- **Platform**: GitHub Actions
- **Registry**: [container registry]

## Approved Variances

| Service | Variance | Justification |
|---------|----------|---------------|
| ml-service | Python | ML ecosystem requirements |
| analytics | MongoDB | Time-series workload |

## Shared Libraries

| Library | Purpose | Version | Used By |
|---------|---------|---------|---------|
| @org/auth-client | Auth service client | 2.x | billing, gateway, notifications |
| @org/logging | Structured logging | 1.x | all services |

7.7 Synthesize draft/.ai-context.md (if not exists or is placeholder)

After generating draft/architecture.md, derive a condensed draft/.ai-context.md using the Condensation Subroutine (as defined in core/shared/condensation.md). This provides a token-optimized, self-contained AI context file at the root level aggregating all service knowledge.

• Read the synthesized draft/architecture.md
• Condense into 200-400 lines covering: system overview, service catalog, inter-service dependencies, shared infrastructure, cross-cutting patterns, critical invariants, and entry points
• If draft/.ai-context.md already exists and is not a placeholder, prompt before overwriting

Step 8: Create Root Config

Create draft/config.yaml if not exists:

# Draft Index Configuration

# Service detection patterns (immediate children only)
service_patterns:
  - "package.json"
  - "go.mod"
  - "Cargo.toml"
  - "pom.xml"
  - "build.gradle"
  - "pyproject.toml"
  - "requirements.txt"
  - "Dockerfile"

# Directories to exclude from scanning
exclude_patterns:
  - "node_modules"
  - "vendor"
  - ".git"
  - "draft"
  - ".*" # Hidden directories

# Re-index on these events (for CI integration)
reindex_triggers:
  - "service added"
  - "service removed"
  - "weekly"

Step 8.5: Refresh Graph Injection Slots

For each initialized service with both draft/architecture.md AND draft/graph/schema.yaml:

A. Read current architecture.md into memory.

B. Regenerate slot content from graph JSONL:

• GRAPH:module-deps → run scripts/tools/mermaid-from-graph.sh --repo . --diagram module-deps Parse JSON response, extract .mermaid string + filtered flag + stats
• GRAPH:proto-map → run scripts/tools/mermaid-from-graph.sh --repo . --diagram proto-map Parse JSON response, extract .mermaid string + stats
• GRAPH:hotspots → read draft/graph/hotspots.jsonl, build top-10 markdown table: | File | Lines | fanIn | Score | with one row per hotspot, ordered by score descending

C. For each slot, find <!-- GRAPH:{id}:START --> ... <!-- GRAPH:{id}:END --> markers. Replace entire block (inclusive of markers) with regenerated content. If a marker pair is absent (legacy file): insert slot at the designated position and log: "Injected GRAPH:{id} slot into architecture.md (slot was absent — legacy file upgraded)"

D. Write updated architecture.md back to disk. Update frontmatter: generated_by = "draft:index", generated_at = now. Also update draft/metadata.json via git-metadata.sh --project-metadata --generated-by "draft:index" to re-anchor synced_to_commit.

E. Re-run Condensation Subroutine (condensation.md) to propagate updated hotspot data into .ai-context.md GRAPH:HOTSPOTS and recompute tier budget. If .ai-profile.md exists, regenerate via Profile Generation Subroutine.

F. Report per service:

✓ <service>: refreshed 3 graph slots (module-deps, proto-map, hotspots)
✓ <service>: regenerated .ai-context.md (tier N, {lines} lines)

Step 9: Completion Report

Remove the lockfile:

rm -f draft/.index-lock

---
                    DRAFT INDEX COMPLETE
---
Scanned: X service directories (depth=1)
Indexed: Y services with draft/ context
Skipped: Z uninitialized services

Generated/Updated:
  ✓ draft/service-index.md (service registry)
  ✓ draft/dependency-graph.md (inter-service topology)
  ✓ draft/tech-matrix.md (technology distribution)
  ✓ draft/product.md (synthesized product vision)
  ✓ draft/architecture.md (system-of-systems view)
  ✓ draft/tech-stack.md (org standards)
  ✓ draft/config.yaml (index configuration)

Service manifests updated: Y services

Next steps:
1. Review synthesized files in draft/
2. Edit draft/product.md to refine overall vision
3. Edit draft/architecture.md to add cross-cutting context
4. Run /draft:index periodically to refresh

For uninitialized services, run:
  /draft:index init-missing
---

Operational Notes

What This Command Does NOT Do

• No deep code analysis — Reads only existing draft/*.md files
• No source code scanning — That's /draft:init's job per service
• No recursive scanning — Depth=1 only, immediate children
• No duplication — Root files link to service files, not copy content

When to Re-Run

• After running /draft:init on a new service
• After significant changes to service architectures
• Weekly/monthly as part of documentation hygiene
• Before major cross-service planning

Preserving Manual Edits

When regenerating, the skill:

1. Reads existing root files
2. Identifies manually-added sections (not marked as auto-generated)
3. Preserves those sections while updating auto-generated parts
4. Sections between <!-- MANUAL START --> and <!-- MANUAL END --> are never overwritten

Graph injection slots (<!-- GRAPH:...:START --> / <!-- GRAPH:...:END -->) are ALWAYS overwritten during refresh — they are auto-managed. Never place manual content between these markers. Use <!-- MANUAL START --> / <!-- MANUAL END --> for content you want preserved near a slot.