GitHub pull request automation without writing YAML · Talyn

GitHub integration

Write the rule once. It runs on every pull request.

The GitHub integration runs automations on pull request events: on these events, matching these conditions, do these actions. Label it, request a reviewer, post a comment, send an agent to fix it, or drop it in the merge queue. It watches every pull request in your repositories (including ones you did not open) and it runs with the app closed.

The decision you keep making

Every time a pull request touches the migrations directory you add the database reviewer. Every time CI fails on a dependency bump you send an agent at it. Every time something is approved and green you queue it. None of these are decisions any more: they are a rule you are executing by hand because nowhere holds it.

An automation is that rule, written down once, in the app. Open Integrations, choose GitHub, and add it under Automations. No YAML, nothing committed to the repository, and nothing that needs the app to be open.

It reads the event, not your list

A GitHub automation fires off the webhook payload, not off Talyn's own list of tracked pull requests. This is a deliberate choice. It makes the feature worth having: a rule applies to every pull request in a repository you have connected, including ones you did not open and ones nobody has looked at yet.

Where the event does not carry a fact (an issue comment payload describes an issue, so it has no base branch) a condition on that fact fails rather than quietly passing. A gate that passes when it cannot see is not a gate.

What a rule can do

The actions are the things you would otherwise do by hand, plus the things only Talyn can do.

The guards, because automation that loops is expensive

Two of them. Talyn ignores events its own GitHub App produced, so a rule that adds a label does not then fire on the label it just added. And each pull request has a ceiling on how many automation runs it can trigger in an hour, defaulting to five, with the refusal announced once rather than silently.

A rate-limited action is parked and retried when the limit clears, not lost, and the retry re-runs only the actions that had not already succeeded, so a rule whose comment posted and whose label was gated does not comment twice.

Merging is the queue, deliberately

The merge action adds the pull request to Talyn's merge queue rather than calling merge directly. A direct merge on a gated base branch just fails, and a rule that fails silently once a week is worse than no rule. Going through the queue means the pull request waits properly, gets fixed if it breaks, and lands when it is genuinely ready.

Questions, answered.

What are GitHub automations?

Rules you set up once in the GitHub integration. They run on their own: when this happens on a pull request, do these things. Pick the trigger, narrow it with conditions, and pick the actions. They watch every pull request in the repositories you connect, including ones you did not open, and they keep running with the app closed.

Do they need anything committed to my repository?

No. An automation lives in Talyn, not in your repository. There is no YAML file, nothing to review, and turning one off does not need a pull request.

Can an automation run one of my skills?

Yes, as long as the skill is one the backend can read: committed in the repository or saved to your workspace. Skills that only exist in ~/.claude/skills on your machine are refused when you save the rule, because the server cannot see your disk.

Is there a limit on how many automations I can have?

No. You can define as many automations as you want on any plan. What costs box-hours is an automation that runs a skill or a prompt, because that starts an agent on a box. Labels, comments and reviewers cost nothing.

Talyn does this for you.

Coding agents in cloud sandboxes, from the first prompt to the merged pull request, on the Claude or ChatGPT plan you already pay for. Free with 10 box-hours a month.