Litmus Deploy Marker Action

Post a deploy marker to Litmus after a successful deploy. The dashboard renders markers as vertical dashed lines on the behavioral-rate charts, so a shift in abandon, edit, or regen rate lines up with the exact release that caused it.

Usage

- uses: trylitmus/deploy-marker-action@v1
  continue-on-error: true
  env:
    LITMUS_AUTH_TOKEN: ${{ secrets.LITMUS_AUTH_TOKEN }}
    LITMUS_SERVICE: my-service
  with:
    environment: ${{ github.ref_name }}

Setup

1. In the Litmus dashboard, create a secret key (ltm_sk_live_*) with the deploys:write scope.
2. Add it to your repo as the LITMUS_AUTH_TOKEN secret.
3. Drop the action into your deploy workflow after the deploy step succeeds.

Your SDK key is a publishable key (ltm_pk_*) and cannot create markers. This is intentional — publishable keys are shipped in bundles.

Environment variables

Variable	Required	Description
LITMUS_AUTH_TOKEN	yes	Litmus secret key (ltm_sk_*) with deploys:write scope
LITMUS_SERVICE	yes	Service name, written to metadata.service so multiple services on one project can be distinguished
LITMUS_URL	no	Override the ingest URL. Defaults to https://ingest.trylitmus.app

Inputs

Input	Required	Default	Description
environment	no	production	Deploy environment
version	no	github.sha[:7]	Version string shown on the chart
description	no	<service> <environment> deploy	Human-readable label

Multi-service example

One repo, two services sharing a Litmus project — tell them apart via LITMUS_SERVICE:

- uses: trylitmus/deploy-marker-action@v1
  continue-on-error: true
  env:
    LITMUS_AUTH_TOKEN: ${{ secrets.LITMUS_AUTH_TOKEN }}
    LITMUS_SERVICE: api
  with:
    environment: ${{ github.ref_name }}

- uses: trylitmus/deploy-marker-action@v1
  continue-on-error: true
  env:
    LITMUS_AUTH_TOKEN: ${{ secrets.LITMUS_AUTH_TOKEN }}
    LITMUS_SERVICE: worker
  with:
    environment: ${{ github.ref_name }}

Behavior

• Skips quietly (warning annotation, exit 0) when LITMUS_AUTH_TOKEN is empty — safe to merge before the secret is wired up.
• curl --connect-timeout 10 --max-time 30 --retry 2 so a flaky endpoint never blocks your deploy.
• Pair with continue-on-error: true if you want marker failures to never fail the job.
• Payload built with jq --argjson so metadata.run_number is a real JSON number, and commit messages / refs can't inject into the body.

Pinning

• @v1 tracks the latest v1.x.x release (recommended)
• @v1.0.0 pins to an exact release
• @main tracks the latest commit (not recommended for prod)

License

MIT