# Release Agent

> Release a new version of the Observe Agent. Use when the user asks to release, tag, publish, or cut a new version of the observe-agent, or mentions the release process.

- Skill: `observeinc/release-agent` (Agent Skill)
- Install (CLI): `npx skillmds@latest add observeinc/release-agent`
- Raw SKILL.md: https://api.skillmd.com/api/skills/observeinc/release-agent/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: observeinc (https://skillmd.com/u/observeinc)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/observeinc/release-agent

---


# Release Observe Agent

## Step 1: Review changelog and determine version bump

Run the [Generate Release Changelog](https://github.com/observeinc/observe-agent/actions/workflows/generate-changelog.yaml) action if not already done. Then determine the latest released version and review changes:

```bash
git tag --sort=-v:refname | head -10
git log <last-tag>..HEAD --oneline
git log <last-tag>..HEAD --oneline --grep="!"  # breaking changes
```

Present the changes to the user and recommend a version bump:

- **Major**: Breaking changes — removed metrics/logs/traces, removed components,
  removed CLI commands/flags, removed OS/arch support, major OTel Collector upgrade
- **Minor**: Non-breaking features — new metrics/logs/traces, new components,
  new CLI commands/flags, OTel minor/patch upgrades, sampling/filtering changes
- **Patch**: Security fixes, smaller bugfixes

Confirm the new version number with the user before proceeding.

## Step 2: Check for CVEs

First, do a quick local check by building the Docker image and running Docker Scout:

```bash
goreleaser build --snapshot --id=linux_build --skip=validate --single-target
cp dist/linux_amd64/linux_build_linux_amd64_v1/observe-agent .
docker build --build-arg TARGETPLATFORM=. -f packaging/docker/Dockerfile -t observeinc/observe-agent:test .
docker scout cves --only-fixed --only-severity medium,critical,high observeinc/observe-agent:test
```

If the local scout check is clean, proceed to the next step. If CVEs are found (or to get full recommendations), also run the [Docker Image Vulnerability Check](https://github.com/observeinc/observe-agent/actions/workflows/vuln-check-full.yaml) against `main`. If critical or high vulnerabilities are found:

1. Identify the import chain for the vulnerable package:

```bash
go mod why -m $PACKAGE_NAME
```

2. If the package comes from `otel-collector-contrib`, check if the latest
   upstream version resolves the CVE. If so, bump versions in
   `builder-config.yaml` — note that all OTel versions (base collector + components)
   usually need to be bumped together.

3. If the package is imported directly, run:

```bash
go mod tidy
```

Repeat until no critical/high CVEs remain.

## Step 3: Test with a nightly release

Trigger the [Release Nightly Version](https://github.com/observeinc/observe-agent/actions/workflows/release-nightly.yaml) action. This verifies:
- The goreleaser build process succeeds
- The built Docker image works correctly

The nightly image is automatically deployed to the eng cluster. You can also
test locally using the [agent dev-tools repo](https://github.com/observeinc/observe-agent-dev-tools/blob/main/snippets.md#agent-helm):

```bash
./start_k8s_demo.sh -f helm/values/nightly-image.yaml
```

Verify via the **Observe MCP** that all expected data from the **eng cluster** is flowing:
- **k8s explorer**: OTel demo pods should appear as running pods
- **APM / traces**: OTel demo services should appear with traces and logs
  (if APM transform is running slowly, check traces directly)

Do not proceed to tagging until nightly testing passes.

## Step 4: Create and push the tag

Ensure you're on `main` with a clean working tree, then:

```bash
git tag v<new-version>
git push origin tag v<new-version>
```

## Step 5: Monitor the Release Version action

The tag push triggers the [Release Version](https://github.com/observeinc/observe-agent/actions/workflows/release.yaml) workflow.

**You MUST actively monitor the workflow until it reaches a terminal state
(success or failure).** Do not leave this step to the user — poll the run
status in the background and report the outcome.

1. Find the triggered run:

```bash
gh run list --repo observeinc/observe-agent --workflow release.yaml --limit 1 --json databaseId,status,conclusion
```

2. Watch it until completion (past successful releases take ~30-37 minutes):

```bash
gh run watch <run-id> --repo observeinc/observe-agent
```

3. On failure, fetch logs and surface them:

```bash
gh run view <run-id> --repo observeinc/observe-agent --log-failed | tail -100
```

Do not proceed to Step 6 until the workflow succeeds.

## Step 6: Draft release announcement

Once the action succeeds, draft a message for the **#observe-agent** slack channel
with this structure:

```
Observe Agent v<version> Released :rocket:

We've just released v<version>! Here's what's new:

**Features**
- <list of feat: commits with PR numbers>

**Fixes**
- <list of fix: commits with PR numbers>

**Infrastructure** (if applicable)
- <list of chore: commits summarized>
```

Include a link to the GitHub release page.