tekton-cel-playground

A small playground that mirrors the behavior of Tekton Triggers' cel-eval: paste a raw HTTP request and a CEL expression, hit Evaluate, and see the result.

The CEL environment exposes the same bindings Tekton Triggers does (see docs):

Binding	Type	Notes
body	dyn (parsed JSON body)	null when the body is not valid JSON
header	map<string, list<string>>	Keys are lower-cased for case-insensitivity
requestURL	string	Path + query from the request line
method	string

Helper member functions:

• header.match(key, value) — true if any value of key (case-insensitive) equals value.
• header.canonical(key) — first value of key (case-insensitive), or "".

Run it

pnpm install
pnpm dev

pnpm dev runs both the Go API (:3002, hot-reloaded by nodemon) and the Next frontend (:3000, proxies /api/evaluate to the Go server) in parallel. Open http://localhost:3000 and pick one of the example presets.

Other scripts:

Script	What it does
pnpm dev	Both servers, both hot-reloaded
pnpm dev:web	Only Next.js dev server
pnpm dev:api	Only Go API, watching *.go and go.mod
pnpm dev:api:once	Run the Go API once with go run main.go
pnpm build	Production build of the frontend

Try it from curl

curl -sS -X POST http://localhost:3002/api/evaluate \
  -H 'Content-Type: application/json' \
  -d '{
    "celExpression": "header.match(\"x-github-event\", \"push\") && body.repository.name == \"hello\"",
    "httpRequest": "POST /webhook HTTP/1.1\nContent-Type: application/json\nX-GitHub-Event: push\n\n{\"repository\":{\"name\":\"hello\"}}"
  }'

Content-Length is recomputed server-side, so you can edit the body without counting bytes.

Deploy to Vercel

The repo is set up for Vercel out of the box:

• src/app/ is the Next.js frontend (auto-detected).
• api/evaluate.go is a Vercel Go serverless function exposed at /api/evaluate — it shares its evaluator with the local server via internal/celeval.
• next.config.ts's rewrite to localhost:3002 is dev-only (gated on NODE_ENV), so in production both routes are served by Vercel.

# one-off
pnpm dlx vercel        # link the project, then 'vercel deploy'
pnpm dlx vercel --prod # production deploy

Or push to a Git remote and connect the repo in the Vercel dashboard — no extra configuration needed. The Go runtime is detected automatically from api/*.go; Vercel will run go mod download and build the function.

First request after idle has a ~300–500 ms cold start (cel-go init). For always-warm latency, deploy the Go server separately on Fly.io / Railway and point the dev rewrite (or production rewrite) at its public URL.

Project layout

api/evaluate.go         Vercel serverless function (production)
main.go                 Local dev HTTP server
internal/celeval/       Shared parser + CEL evaluator
src/app/                Next.js frontend

Contributing

Check CONTRIBUTING.md for the steps.