Repo Doc Generator — VS Code Extension
One click: generate AI-powered branch documentation, published to Confluence.
Switch to a feature branch → hit Cmd+Shift+G → get a technical review doc AND a non-technical summary in Confluence. Run again on the same branch and the same pages get updated.
Quick Start
- Install the extension (or F5 from source)
- Open a git repository in VS Code
- Setup wizard walks you through LLM + Confluence configuration
- Switch to a feature branch
Cmd+Shift+G (or Command Palette → "Repo Doc: Run")- Two docs appear in Confluence under your parent page
How It Works
You click "Run"
│
├── Step 1: Detect branch, compute diff vs main
├── Step 2: Scan repo structure (languages, key files)
├── Step 3: Generate technical + non-technical docs via LLM
├── Step 4: Publish to Confluence (create or update)
└── Step 5: Done — links shown in notification
The extension tracks branch → page ID mapping. Same branch = same pages updated (version incremented), never duplicated.
Configuration
All settings under repoDoc.*:
| Setting |
Default |
Description |
llm.provider |
anthropic |
anthropic, openai, or bedrock |
llm.model |
claude-sonnet-4-20250514 |
Model ID for your provider |
confluence.baseUrl |
— |
Your Confluence URL |
confluence.spaceKey |
— |
Target space key |
confluence.parentPageId |
— |
Parent page for generated docs |
baseBranch |
main |
Branch to compare against |
docLength |
concise |
concise / standard / detailed |
bedrock.region |
us-west-2 |
AWS region (Bedrock only) |
bedrock.profile |
— |
AWS SSO profile (Bedrock only) |
Secrets (API keys, Confluence token, email) are stored in VS Code's SecretStorage (OS keychain).
Commands
| Command |
Shortcut |
Description |
| Repo Doc: Run |
Cmd+Shift+G |
Generate/update branch docs |
| Repo Doc: Setup |
— |
Re-run setup wizard |
| Repo Doc: View Docs |
— |
Open last generated docs |
Development
npm install
npm run compile
# F5 to launch Extension Development Host
npm run watch # continuous compilation
Architecture
src/
├── extension.ts — Entry point, command registration, orchestration
├── config/ — Setup wizard (webview), settings, SecretStorage
├── git/ — Branch detection, three-dot diff
├── scanner/ — Repo file walker, language detection
├── llm/ — Provider factory (Anthropic, OpenAI, Bedrock)
├── generator/ — Prompt templates, parallel doc generation
├── publisher/ — Confluence REST API (create/update/version)
└── tracker/ — Branch → page ID mapping (workspaceState)
Supported Providers
| Provider |
Auth |
Models |
| Anthropic |
API Key |
claude-sonnet-4-20250514, claude-opus-4 |
| OpenAI |
API Key |
gpt-4o, gpt-4-turbo |
| AWS Bedrock |
IAM/SSO |
Any Bedrock-hosted model (ARN) |
Finding Your Confluence Parent Page ID
- Navigate to the target parent page
- Click "..." → "Page Information"
- The page ID is in the URL:
.../pages/viewinfo.action?pageId=12345
Security & Data Flow
This extension is a pipe, not a platform. Your code goes directly from your machine to the APIs you configure — no middleman, no data passes through any server we control.
Your machine (git diff) ──► LLM API (your creds) ──► back to your machine ──► Confluence (your instance)
What we do NOT do
- No intermediary server — zero network calls except to APIs you explicitly configured
- No telemetry that captures code content
- No persistent storage of diffs or generated docs on disk
- No "phone home", analytics, or usage tracking
Who owns what
| Concern |
Owner |
Details |
| Code sent to LLM |
You |
You chose the provider and entered your own API key |
| LLM data retention |
Your LLM provider |
Anthropic/OpenAI API: don't train on API data. Bedrock: stays in your AWS account |
| Docs in Confluence |
You |
Your instance, your space, your credentials |
| API keys at rest |
VS Code SecretStorage |
OS-level encryption (Keychain on macOS, Credential Manager on Windows, libsecret on Linux) |
For maximum data isolation: use AWS Bedrock
With Bedrock, your code never leaves your own AWS account. The LLM call is made within your VPC boundary. This is the recommended option for enterprise teams with strict data residency requirements.
Code-level safety
- All credentials in OS keychain (SecretStorage) — never in plaintext settings.json
- Publishes only to your configured space + parent page
- Git commands use
execFile (immune to shell injection)
- No telemetry or external reporting
- No third-party dependencies that phone home
