CodeHerderSearch⌘KRequest access →

The commits and merge requests a task produced

Find the commits and merge requests one task made, from the task page's panels or from `ch task commits` and `ch task merge-requests`.

A task finishes a stage, and the next question is always the same: where’s the code? CodeHerder keeps a record of every commit and every merge request a task’s sessions produced, and you can read it from the web app or the CLI.

On the task page

Open a task and look for two panels: Merge requests, subtitled “opened by this task’s sessions”, and Commits, subtitled “made by this task’s sessions”. Each lists rows in the same shape:

  • Merge requests — Number, Title, Repo, State, and when it was last updated. State is Merged with the merge time, or Unmerged. An Awaited badge marks the requests the task waits on.
  • Commits — SHA, commit message, author, repo, and when it was committed. For who the author is, see Whose name is on a commit.

The SHA and the merge-request number link out to the provider’s own page, when CodeHerder has resolved one. The Repo cell links to that repo’s page inside CodeHerder. Before anything’s landed, each panel shows an empty state — “No merge requests recorded yet.” or “No commits recorded yet.” — rather than an error.

A session page carries the identical pair of panels, scoped to just that one session’s sandbox instead of the whole task.

Whose name is on a commit

Each commit has two names: an author and a committer. CodeHerder sets both when it starts an agent session, and a launch configuration can’t override them. The one exception is a repo with its own commit identity.

  • Committer — the agent. Its name is the agent’s display name followed by “(CodeHerder agent)”, for example “Ada (CodeHerder agent)”. The email is a placeholder address that ends in @codeherder.invalid and can’t receive mail. A commit made with no session identity shows the name “CodeHerder agent”.
  • Author — the person who filed the original task. CodeHerder follows “filed by” links up from the task to the first task a person filed, and uses that person. If it finds none, it uses the task’s creator when that’s a person. The name is their display name, or their email address if they have no display name.

That person needs a verified email address. If they don’t have one, or no person filed any task in the chain (for example, an integration or webhook filed it), the agent is both author and committer.

A repo can carry a commit identity. When it does, an agent’s commits in that repo use that name and email as both author and committer, and the rules above don’t apply to that repo. Other repos in the same session keep them. A workspace owner or admin sets it with the CLI; see Setting a commit identity for a repo.

In a session you start on a device as an agent, you’re the author if your email address is verified, because you started it. The agent is still the committer. See Start a session on a device.

To see both names on a commit in a checkout:

git log --format='%h  author: %an  committer: %cn'
3f9c2ab  author: Sam Rivera  committer: Ada (CodeHerder agent)

A few limits to know:

  • The merge request isn’t in the agent’s name. Your git host shows it as opened by the owner of the git credential the device used to push it: the GitHub App’s [bot] account for a minted token, or the token’s owner for a pooled one. See Git credentials on a device.
  • The author line is attribution, not proof. Commits aren’t signed, and an agent process can change the identity it commits with. Rely on review and branch protection, not on the name. See Reviewing an agent’s work.
  • Older commits can differ. Commits from sessions that started before this behaviour shipped can show the agent as author.

From the CLI

ch task commits <taskId>
ch task merge-requests <taskId>

ch task commits lists every commit the task’s sessions made, across every sandbox the task worked in, newest first. ch task merge-requests lists every merge request those sessions produced, the same way.

Both lists are capped and don’t page — unlike most ch list commands, neither takes --limit or --cursor. A task’s own git history is small enough that CodeHerder just hands you the whole thing.

Inside an agent session, <taskId> defaults from CH_TASK_ID, so the bare form works without naming it:

ch task commits
ch task merge-requests

Narrowing to one sandbox

A task can attach more than one sandbox over its lifetime — a rework pass gets its own, for instance. To see just one sandbox’s slice instead of the whole task’s, use the sandbox-scoped equivalents:

ch sandbox commits <sandboxId>
ch sandbox merge-requests <sandboxId>

These skip --limit/--cursor too. Inside an agent session, <sandboxId> defaults from CH_SANDBOX_ID the same way <taskId> defaults from CH_TASK_ID, so the bare form works there too.

The requests a task waits on

ch task merge-requests shows one list. Besides every merge request any of the task’s sessions opened, it marks the ones the task waits on before it counts as landed, one per repo, in the AWAITED column, the same requests that carry the panel’s Awaited badge. See How the work lands for recording one, and Reviewing an agent’s work for the merge checkpoint itself.

Why a list can be incomplete

A merge request lands in these lists one of three ways: an agent reports it the moment it opens one, someone records it with ch task merge-requests create (or an agent writes its link into artifact), or a periodic sweep finds it by matching branches against your git host. The first two show up right away and need nothing configured. The sweep needs a git-host integration connected for that repo’s host (see Integrations) — without one, CodeHerder has no way to look up merge requests it wasn’t told about directly, and only the reported or recorded ones appear.

Last updated

Try it on one repository for two weeks.

File a task. An agent builds it. A person reads every request. You see the cost of each task.

Try "pricing", "connect a device", or "who reviews the code"

↑↓ move · ↵ open · esc close