Release Lambda Layer Skill for Claude

Release a new datadog-lambda-js Lambda layer

This mirrors the "Release" section of the SLS Confluence page "Lambda Layer

Node.js" (Commercial + GovCloud). It assumes a Release Candidate has

already been validated (self-monitoring + e2e tests passed) before this

process starts.

This process pushes git tags, triggers production-publishing GitLab jobs,

publishes to the public npm registry, and creates a public GitHub release.

Treat every step in the "Guardrails" section as a hard stop — confirm

with the user before running it, even if they've approved a similar step

earlier in the same run.

Before starting

• Announce the start of the release in a #apm-serverless Slack thread.

Use this same thread for every update below. Ask the user to post, don't

post on their behalf unless they've explicitly asked you to send Slack

messages for them.

• Confirm a Release Candidate deploy already passed self-monitoring and

e2e tests for this change. Do not start Phase 1 without it.

Phase 1 — Commercial release

Reference: https://datadoghq.atlassian.net/wiki/spaces/SLS/pages/3375925277

• Create a branch for the version bump.
• Run yarn upgrade dd-trace (or yarn upgrade dd-trace@^a.b.c for a

specific version) to pick up the latest tracer.

• If dd-trace changed, refresh integration test snapshots:

BUILD_LAYERS=true UPDATE_SNAPSHOTS=true DD_API_KEY=<key from 1Password> aws-vault exec sso-serverless-sandbox-account-admin -- ./scripts/run_integration_tests.sh

• Bump package.json version to X.Y.0:
• Minor version bumps on every release, and **must match the layer

version**.

• Bump major too for breaking changes.
• Commit as vX.Y.0, open a PR, get approval.
• Before merging, confirm a Release Candidate deploy already passed

for this change (per "Before starting" above).

• Merge the PR. **Tell the user not to merge any other PR until this

release finishes** — a concurrent merge can tag the wrong commit.

• git checkout main && git pull
• git tag vX.Y.0

10. git push --tags origin main — **guardrail: confirm with the user

before running this.** It kicks off the production build in GitLab and

can't be cleanly undone.

11. In GitLab tags,

open the pipeline for the new tag.

12. Guardrail: confirm before triggering, then run the `sign-layer

(nodeXX)` manual job for every supported runtime. This deploys the

layer to every commercial region.

• Claude does not have GitLab UI/API access by default, so this is

normally an instruction the user runs themselves — walk them through

clicking the job in the GitLab pipeline and wait for them to confirm

it finished before moving on. Only trigger it directly if the user

has explicitly set up GitLab API credentials for Claude to use, and

still confirm before every single trigger, per the guardrail above.

13. Guardrail: confirm before triggering, then run the

publish-npm-package manual job (same access caveat as step 12).

Afterward, verify publicly:

    mkdir new-dir && cd new-dir && npm init -y
    npm install datadog-lambda-js@<new version>

14. Guardrail: confirm before creating, then draft a GitHub release on

the releases page:

• Select the new tag and the previous tag, generate release notes.
• Call out which dd-trace version is packaged in this release (from

step 2 of Phase 1) — this is important for users debugging tracer

behavior against a specific layer version.

• Paste in layer ARNs for every runtime and region pattern, e.g.:

      arn:aws:lambda:<AWS_REGION>:464622532012:layer:Datadog-Node18-x:117
      arn:aws:lambda:<AWS_REGION>:464622532012:layer:Datadog-Node20-x:117
      arn:aws:lambda:<AWS_REGION>:464622532012:layer:Datadog-Node22-x:117

      arn:aws-us-gov:lambda:us-gov-<AWS_REGION>:002406178527:layer:Datadog-Node18-x:117
      arn:aws-us-gov:lambda:us-gov-<AWS_REGION>:002406178527:layer:Datadog-Node20-x:117
      arn:aws-us-gov:lambda:us-gov-<AWS_REGION>:002406178527:layer:Datadog-Node22-x:117

(substitute the actual layer version number for 117)

• Attach the layer zip files.

15. Post the finished release + release notes link to the Slack thread.

Phase 2 — GovCloud (manual)

No automated pipeline exists for this — compliance requirements mean it has

to stay manual.

• Confirm AWS SSO is set up for both Commercial and GovCloud (see

cloud-inventory setup).

• Check the Phase 1 pipeline for a single job that bundles all runtimes'

signed layers together (e.g. a "signed layer bundle" artifact) — if one

exists, download that instead of going job-by-job. Otherwise, fall back

to downloading each individual sign-layer (nodeXX) job's artifacts.

Place the resulting layer bundle zips in the local .layers folder.

• Guardrail: confirm before running, then for each GovCloud

environment:

   VERSION=<LAYER_VERSION> ENVIRONMENT=us1-fed ./scripts/publish_govcloud_layers.sh <path-to-layer-bundle.zip>
   VERSION=<LAYER_VERSION> ENVIRONMENT=us2-fed ./scripts/publish_govcloud_layers.sh <path-to-layer-bundle.zip>

VERSION is the integer layer version (e.g. 116), not the npm semver.

When prompted for auth, the user must open the link in their GovCloud

browser profile — don't attempt to open or approve it yourself.

• Guardrail: confirm before editing, then update

latest-lambda-layer-version.html

in the documentation repo to the new version, via a PR.

Guardrails (always confirm before)

• git push --tags origin main
• Triggering any GitLab manual job that publishes/signs a layer or

publishes the npm package

• Creating the GitHub release
• Running publish_govcloud_layers.sh
• Opening a PR against the documentation repo
• Merging any release-version-bump PR

Failure modes

• If integration test snapshots don't update correctly, check the

installed serverless framework is v3, not v4:

npm install -g [email protected].

• If a teammate merges a PR to main between step 7 and step 9 of Phase 1,

stop and confirm with the user which commit should actually be tagged.