gh-summary — Slack PR Summary Bot

Warning

This is a simple POC. The code hasn't been audited and there are no warranties it will be suitable for produciton environments.

A Slack slash command that fetches a GitHub PR's metadata, CI status, and discussion thread, then produces an AI-powered summary using Claude.

(I've also included a skill with the same goal - you can find it under ./skills).

For another in this series of integration, have a look at The Coroner - a tool to generate post-mortems from a slack channel.

/gh-summary https://github.com/owner/repo/pull/42

Architecture

User (Slack)
    │  /gh-summary <PR URL>
    ▼
Slack Platform ──────────────────────────────────────────────┐
    │  HTTP POST (signed)                                    │
    ▼                                                        │
Provider Adapter Layer                                       │
    ├──► Vercel: waitUntil() keeps function alive             │
    └──► AWS Lambda: Fire-and-forget (CloudWatch logs)        │
    │                                                          │
    ▼                                                          │
Core Business Logic (src/)                                   │
    ├──► src/clients/github.js   — GitHub API client          │
    ├──► src/clients/anthropic.js — Claude AI summarization   │
    ├──► src/clients/slack.js     — Signature verification    │
    ├──► src/blocks/metadata.js   — CI, reviewers, metadata    │
    └──► src/blocks/summary.js    — Text splitting             │
    │                                                          │
    ├──► GitHub REST API  (parallel)                          │
    │       • GET /pulls/:number              (PR metadata)   │
    │       • GET /issues/:number/comments    (discussion)    │
    │       • GET /pulls/:number/reviews      (code reviews)  │
    │       • GET /pulls/:number/comments     (review comments)│
    │       • GET /pulls/:number/files        (changed files) │
    │                                                          │
    ├──► GitHub REST API  (sequential — needs PR head SHA)    │
    │       • GET /commits/:sha/check-runs    (CI status)     │
    │                                                          │
    └──► Anthropic Messages API  (after all GitHub data is ready)
            • claude-sonnet-4-6 (discussion → structured summary)
                │
                ▼
         POST response_url → Slack (formatted Block Kit message)

Why a server at all?

Slack slash commands work by POSTing to your URL. There is no way to point a slash command directly at GitHub or any other third-party API — Slack needs an endpoint it can reach that can verify the request, hold your credentials, and format the response.

Choose your provider:

Vercel Hobby (free) — ~1s function time, simplest setup
AWS Lambda — Pay-per-use, scales with traffic, no cold start limits

What the Bot Returns

Each /gh-summary invocation posts a message with:

Section	Details
Header	PR number and title
Metadata	Status, author, created/updated/merged dates, diff size
Reviewers	Who has been requested to review
CI Status	Pass/fail/pending counts, names of failing checks
Discussion Summary	AI-generated bullet points covering debates, decisions, open questions, and agreed strategies
Link	One-click button to open the PR

Prerequisites

Requirement	Where to get it
Node.js ≥ 18	nodejs.org
Vercel account (free) or AWS account	vercel.com / aws.amazon.com
Slack workspace (admin)	Your workspace settings
GitHub account	github.com/settings/tokens
Anthropic account	console.anthropic.com

Choosing a Provider

Vercel — simplest, free tier sufficient for most use cases.
AWS Lambda — pay-per-use, better for high-traffic or cost-sensitive deployments.

See the Deployment Options section below for setup instructions.

Deployment Options

Option 1: Deploy to Vercel (Recommended for most users)

# Clone / enter the project
cd gh-summary

# Install dependencies
npm install

# Log in to Vercel
npx vercel login

# Deploy (follow prompts — accept all defaults)
npx vercel deploy --prod

Copy the deployment URL, e.g. https://gh-summary-xxx.vercel.app

Option 2: Deploy to AWS Lambda + API Gateway

Install SAM CLI: docs.aws.amazon.com/sam/latest/cli/installation.html
```
pip install aws-sam-cli
sam --version
```

Create secrets in AWS Secrets Manager:

aws secretsmanager create-secret \
  --name gh-summary/secrets \
  --secret-string '{
    "SLACK_SIGNING_SECRET": "your-slack-signing-secret",
    "GITHUB_TOKEN": "your-github-token",
    "ANTHROPIC_API_KEY": "your-anthropic-api-key"
  }'

Deploy with SAM:

sam build
sam deploy \
  --stack-name gh-summary-stack \
  --template-file aws/sam-template.yaml \
  --capabilities CAPABILITY_IAM

Get your API Gateway URL from the deployment output, e.g.:

https://abcdefgh123.execute-api.us-east-1.amazonaws.com/prod/gh-summary

Step 5 — Set Environment Variables (Vercel Only)

In your Vercel project dashboard, add:

Variable	Value
`SLACK_SIGNING_SECRET`	From Slack App settings
`GITHUB_TOKEN`	From GitHub token generator
`ANTHROPIC_API_KEY`	From Anthropic console
`PROVIDER`	`vercel` (default)
`CAPPED`	(optional) Set to `true` on Vercel Hobby for timeout safety
`MAX_TOKENS`	(optional) Max tokens for Claude's response. Defaults to `1000`

Or via CLI (Vercel):

npx vercel env add SLACK_SIGNING_SECRET
npx vercel env add GITHUB_TOKEN
npx vercel env add ANTHROPIC_API_KEY
npx vercel env add PROVIDER
npx vercel env add CAPPED
npx vercel env add MAX_TOKENS

# Redeploy to apply env vars
npx vercel deploy --prod

Step 6 — Wire Up Slack

Go back to your Slack App → Slash Commands → edit /gh-summary
Set Request URL to: https://your-project.vercel.app/api/gh-summary (Vercel) or https://your-api-gateway-url/api/gh-summary (AWS)
Save

Step 7 — Test It

In any Slack channel:

/gh-summary https://github.com/vercel/next.js/pull/1

Local Development

# Copy env file and fill in your values
cp .env.example .env

# Start local Vercel dev server (or use Lambda locally with SAM)
npm run dev
# → Listening at http://localhost:3000

# Or test with AWS SAM locally
sam local invoke GhSummaryFunction --event event.json

The test script generates a valid Slack HMAC signature so the local function accepts it.

Security

Signature verification — every request is validated with HMAC-SHA256 against your SLACK_SIGNING_SECRET. Forged or replayed requests are rejected.
Replay protection — requests older than 5 minutes are rejected.
No credential storage — API keys live only in environment variables (Vercel encrypted env vars or AWS Lambda config), never in code.
GitHub token scope — use public_repo unless you need private repo access.

Cost Estimate

Service	Vercel Hobby	AWS Lambda (estimated)
Hosting	Free	~$0.20/month at 100 summaries/day
GitHub API	Free	Free (5000 req/hr limit)
Anthropic AI	~$0.003/summary	Same (~$0.003/summary)

AWS Lambda pricing: $0.400 per 1M requests + $0.0000166667 per GB-second compute.

Troubleshooting

Bot doesn't respond

Check provider logs: npx vercel logs (Vercel) or AWS CloudWatch Logs (Lambda)
Ensure all 4 env vars are set and you redeployed after adding them

"Unauthorized" error

SLACK_SIGNING_SECRET is wrong or missing

GitHub 404 error

For private repos, ensure your token has the repo scope (not just public_repo)

"Could not generate AI summary"

Check your ANTHROPIC_API_KEY is valid and has credits

Slack says "operation_timeout"

The function took > 3s to acknowledge. This shouldn't happen — if it does, check provider cold start times or upgrade plan (Vercel Pro) / increase timeout (Lambda).

AWS Lambda-specific

Check CloudWatch Logs for error messages
Ensure API Gateway is properly configured with CORS headers

File Structure

gh-summary/
├── api/
│   └── gh-summary.js          # Entry point (uses adapter pattern)
├── src/                        # Provider-agnostic core logic
│   ├── adapters/              # Platform-specific implementations
│   │   ├── index.js           # Factory function (createAdapter)
│   │   ├── vercel.js          # Vercel adapter with waitUntil()
│   │   └── lambda.js          # AWS Lambda adapter (fire-and-forget)
│   ├── clients/               # API integrations
│   │   ├── github.js          # GitHub REST client + URL parsing
│   │   ├── anthropic.js       # Claude AI summarization logic
│   │   └── slack.js           # Signature verification helpers
│   └── blocks/                # Block Kit builders
│       ├── metadata.js        # CI status, reviewers, PR metadata
│       └── summary.js         # Text splitting into Slack blocks
├── docs/
│   ├── README.md              # This file
│   └── architecture.svg       # System diagram
├── scripts/
│   └── test-local.js          # Local smoke test
├── aws/                       # AWS deployment (optional)
│   ├── sam-template.yaml      # SAM template for Lambda + API Gateway
│   └── lambda-function.zip    # Deployed function bundle
├── .env.example               # Environment variable template (includes PROVIDER)
├── .gitignore
├── package.json
└── vercel.json                # Vercel configuration

Deployment Options Summary

Feature	Vercel	AWS Lambda
Setup complexity	⭐ Simplest	⭐⭐ Moderate
Cost (low traffic)	Free	~$0.20/month
Cold starts	< 1s typical	1-3s first request
Timeout handling	`waitUntil()` built-in	Fire-and-forget
Switching providers	Change `.env` only	Change `.env` only

The modular architecture means adding a new provider (GCP, n8n, etc.) is as simple as:

Creating src/adapters/gcp.js implementing the same interface
Registering it in src/adapters/index.js
Setting PROVIDER=gcp in .env

Done! 🎉

Known limitations

Vercel Hobby timeout — Vercel Hobby has a 30s max function duration. For large PRs, Claude can take longer than that, causing the response to never reach Slack. Set CAPPED=true to enforce input limits and stay within the 30s timeout (Free tier friendly). On paid plans or AWS Lambda, omit CAPPED (or set it to false) to send full data.
AWS Lambda cold starts — First request after deployment may take 1-3 seconds due to initialization. Subsequent requests are faster (< 200ms typically).

Extending the Bot

Some ideas if you want to go further:

PR labels / milestones — already in the PR API response, just add them to the Block Kit blocks
Support for Vercel alternatives - Adding support for n8n or lambda functions, ideally in a modularized way

Adding a New Provider (e.g., GCP, n8n)

The modular architecture makes this straightforward:

Create src/adapters/gcp.js implementing the same interface as vercel.js and lambda.js:

export class GCPAdapter {
  async acknowledge(responseUrl, payload) { /* ... */ }
  async processPR(prUrl, responseUrl) { /* fire-and-forget */ }
  async postToSlack(responseUrl, payload) { /* ... */ }
}

Register it in src/adapters/index.js:

export function createAdapter(provider) {
  if (provider === 'gcp') return new GCPAdapter();
  // ... existing providers
}

Set PROVIDER=gcp in your .env file and redeploy.

That's it! The core business logic (src/clients/) remains unchanged.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

gh-summary — Slack PR Summary Bot

Contents

Architecture

Why a server at all?

What the Bot Returns

Prerequisites

Choosing a Provider

Deployment Options

Option 1: Deploy to Vercel (Recommended for most users)

Option 2: Deploy to AWS Lambda + API Gateway

Step 5 — Set Environment Variables (Vercel Only)

Step 6 — Wire Up Slack

Step 7 — Test It

Local Development

Security

Cost Estimate

Troubleshooting

File Structure

Deployment Options Summary

Known limitations

Extending the Bot

Adding a New Provider (e.g., GCP, n8n)

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 20 Commits
api		api
aws		aws
docs		docs
lib		lib
scripts		scripts
skills/gh-summary		skills/gh-summary
src		src
tests		tests
.env.example		.env.example
.gitignore		.gitignore
.samignore		.samignore
README.md		README.md
package-lock.json		package-lock.json
package.json		package.json
samconfig.toml		samconfig.toml
vercel.json		vercel.json
vitest.config.js		vitest.config.js

Folders and files

Latest commit

History

Repository files navigation

gh-summary — Slack PR Summary Bot

Contents

Architecture

Why a server at all?

What the Bot Returns

Prerequisites

Choosing a Provider

Deployment Options

Option 1: Deploy to Vercel (Recommended for most users)

Option 2: Deploy to AWS Lambda + API Gateway

Step 5 — Set Environment Variables (Vercel Only)

Step 6 — Wire Up Slack

Step 7 — Test It

Local Development

Security

Cost Estimate

Troubleshooting

File Structure

Deployment Options Summary

Known limitations

Extending the Bot

Adding a New Provider (e.g., GCP, n8n)

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages