release-manager · ruvnet — Claude Code subagent: releases and deploys
release-manager runs a multi-package release: version bumps, tests, deploy and rollback, then keeps what worked in memory for the next one.
no license ★ 850 updated 2026-10-26
.claude/agents/release-manager.mdWhat release-manager does
The agent carries a release from version bumps to the deploy itself. It raises version numbers across packages, runs the test suite, ships, and rolls back when something breaks afterwards. Release notes come out of the same pass. Before shipping it polls several agents and folds their verdicts into one answer with a confidence figure.
Memory sits in the hooks. The pre block queries AgentDB for earlier releases that scored well, using npx agentdb-cli pattern search, and pulls their stats. The post block writes the run back as a pattern with a reward, and on a successful run above 0.9 it retrains a coordination pattern through npx claude-flow neural train. Tools cover shell, file edits, Task, WebFetch, GitHub MCP calls for branches, pull requests and issues, plus claude-flow swarm tools and agentic-flow pattern storage.
Connect the github and claude-flow MCP servers before you rely on it: without them most of the tool list never resolves and only Bash and Edit remain. The hooks also shell out through npx, so Node has to be on the machine.
What it helps with
- bump versions across several packages in one pass
- run the tests and produce release notes
- roll back a deploy that broke after shipping
- turn QA and security verdicts into one go/no-go call
- reuse a past release's pattern instead of repeating its mistakes
Example prompts for the agent
- Cut release 2.0: bump versions in every package, run the tests and open a PR against
main. - Collect the QA and security verdicts for this release and tell me whether we ship.
- Deploy the release and roll back to the previous tag if the healthcheck fails.
How to install release-manager
Manually: put the files in place
.claude/agents/release-manager.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/release-manager.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/release-manager.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about release-manager
How to install release-manager in Claude Code?
Place the agent file at .claude/agents/release-manager.md inside your project.
What does release-manager do?
It coordinates a release across packages: version bumps, tests, deploy with rollback, and release docs. Before shipping it merges several agents' verdicts into one decision.
Why does it need the MCP servers?
GitHub work such as branches, pull requests and issues goes through the github MCP tools, and swarm tasks through claude-flow. Disconnect them and those tools disappear from the agent's reach.
Similar subagents
All subagents for Agent setup: memory, context, orchestration →release-manager by other authors comes first.