tend
An agent that tends your repo — reviewing PRs, triaging issues, and responding to whatever comes up, so you can focus on building beautiful things.
claude plugin marketplace add max-sixty/tend
claude plugin install install-tend@tend
claude /install-tenda small field guide
A few of the things tend does, day to day.
PR reviews
Reads each pull request. It draws on the Skills and docs your team already wrote, so the review weighs in on what your project cares about.
Because it runs in your CI, it can run the tests against the change and iterate on a fix until they pass. It'll offer to push the fix.
Issue triage
When an issue opens, tend classifies it, checks the backlog for duplicates, and tries to reproduce any reported bug.
If the fix is small and conservative, it'll open a PR. Otherwise it leaves a clear note for whoever picks it up.
CI fixes
When the default branch goes red, tend reads the failed run, finds the underlying cause, and opens a PR with the fix.
Work delegation
Mention the bot in a PR or issue and it takes a pass at what you asked — resolve a merge conflict, draft a change, research a question, fix the tests.
Nightly & weekly maintenance
Overnight it sweeps a rotating sample of files for stale docs and small bugs, reviews the day's commits, resolves merge conflicts on its own open PRs, and closes issues that got fixed.
Each week it works through dependency PRs — approving the patch and minor bumps with green CI, flagging the rest.
Self-review
Each day tend reads the previous day's runs of its own workflows, looks for behaviour that didn't sit right, and proposes adjustments to its Skills and config.
recent ground
A few of the last things the bots have touched.
how it works
A subscription, some workflows, and Skills you can shape.
Tend runs Claude Code or Codex in GitHub Actions, funded by your subscription or API key.
In your repo, Tend is a handful of GitHub Actions workflows, each running a Skill, generated from a single config.
Skills and docs in your repo overlay Tend's defaults. Set what it auto-approves, calibrate what it flags in reviews, have it push small fixes itself, close some category of PR on sight, or give it a new chore to run each night.
getting started
Claude Code sets Tend up in about ten minutes.
claude plugin marketplace add max-sixty/tend
claude plugin install install-tend@tend
claude /install-tendThe /install-tend Skill walks the rest of the way: config file, workflow generation, bot account, secrets, branch protection. It'll ask before each step that touches your repo.
You'll need a GitHub account for the bot (something like @your-project-bot) and model access: a Claude subscription or Anthropic API key, an OpenAI API key, or experimental ChatGPT Plus or Pro authentication for Codex. Maintainers of sizeable OSS projects can get a Max subscription free from Anthropic.
security
Built on the assumption that a pull request can take over the agent.
By default, a GitHub ruleset prevents the bot from landing code; a maintainer merges each proposal. Optional yolo mode lets the bot merge ordinary PRs, while workflow, config, and agent-instruction changes still require a maintainer's approval.
The agent runs in a sandbox. Proxies outside it hold the bot PAT and API credentials, and workflows on unprotected branches can't read Tend's operational secrets. Codex subscription sessions receive an expiring access token, but never the refresh token.
Until the run ends, it can still act as the bot and send out what the agent can read. Full threat model →