Repository navigation
t3code-giveback: replacing a limit-closed PR, and a stale PR head behind a stopped-moved stop #8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Changes from all commits
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -60,7 +60,13 @@ Open a `text_approval` ask carrying the exact text. When he approves an issue's | |
| - Rebase onto the latest `origin/main` and re-run the focused tests. Before any push to the branch of a PR that is already open, check its labels: see the landing pipeline in step 6. | ||
| - **Check the posting limits first** (standing rule 7): `giveback capacity` shows the posts of the last 24 hours and 7 days, and the new PRs and issues, each beside its limit. It counts from GitHub, so it includes posts no agent recorded. If `posts.source` is `events`, GitHub could not be read and the counts may be low: check his recent activity on GitHub yourself before posting. If this post would pass a limit, wait, and tell Jackson if it cannot wait. | ||
| - **Then the open PR limit.** Run `gh pr list -R pingdotgg/t3code --author Jacksondr5 --state open`. If 3 are open, don't post: put the item in `held` with `waitingOn: pr_slot`, and tell the dispatcher. This is the one use of `held` that needs no `covering` or `base` link. | ||
| - **Waiting for a slot.** The dispatcher sets the item's `queueRank` in the order Jackson gives. A slot frees when one of our PRs merges or closes. The dispatcher learns that from the watcher, which tells it whenever one of our PRs closes or merges, and from its daily check. It then tells the agent of the first item in the queue, and that agent rebases, re-runs the focused tests, and opens its PR, or reopens the one the bot closed. Never open or reopen before the dispatcher says so. | ||
| - **Waiting for a slot.** The dispatcher sets the item's `queueRank` in the order Jackson gives. A slot frees when one of our PRs merges or closes. The dispatcher learns that from the watcher, which tells it whenever one of our PRs closes or merges, and from its daily check. It then tells the agent of the first item in the queue, and that agent rebases, re-runs the focused tests, and opens its PR. Never open one before the dispatcher says so. | ||
| - **A PR the limit bot closed cannot be reopened by us.** Open a replacement PR in its place: | ||
| - from the same branch, rebased onto `main`; | ||
| - with a body that starts with one line linking the old PR, such as "Replaces #36959, which the open PR limit closed." That line is part of the text Jackson approves; | ||
| - with **"Allow edits by maintainers" on** (`gh pr create` sets it unless told `--no-maintainer-edit`). The landing pipeline pushes to our branch, and a PR it cannot push to gets replaced by one of its own. | ||
| - On the dashboard, the new PR becomes the item's `primary` link and the old one stays as `related`. | ||
| - A replacement is not a new PR under Jackson's limits. It counts as one post in the 24-hour and 7-day totals. The dashboard recognises it by the branch: the PR just before it on the same head branch was closed by the limit bot as it opened. So keep the branch name. | ||
| - Open the PR: `gh pr create -R pingdotgg/t3code --head Jacksondr5:<branch>`. An issue is filed differently: see "Filing an issue" below. | ||
| - In the same session: | ||
| - set the item's `primary` link (`ownership: ours`) and stage `offered`; | ||
|
|
@@ -108,6 +114,7 @@ What we do: | |
|
|
||
| - **While a PR carries `takeover:in-progress` or `takeover:ready`, never push to its branch,** not even a mechanical rebase, and do not reply on it. A push from anyone else stops the pipeline. Check the labels before any push: `gh pr view <n> -R pingdotgg/t3code --json labels`. | ||
| - **After `takeover:blocked`, a closure by the pipeline, or a pipeline comment that asks for steps:** bring Jackson the reason and a proposed reply or fix. He approves it before anything goes up. | ||
| - **A `stopped-moved` stop is not always a real outside push.** On 2026-11-15 the pipeline pushed its merge of `main` to one of our fork branches, GitHub never moved the PR's head to that commit, and the pipeline then stopped with `factory:stopped-moved`, as if someone else had pushed. Before you accept such a stop, compare the fork branch's tip (`git ls-remote j5 refs/heads/<branch>`) with the PR's head (`gh pr view <n> -R pingdotgg/t3code --json headRefOid`). If they differ, the PR's head is stale on GitHub's side. Bring Jackson both SHAs and what you propose. Push nothing to repair it without his say. | ||
|
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Do not treat one mismatch as proof of a stale PR head. If another actor pushes between the two commands, Recheck both values after a mismatch. Describe a persistent mismatch as evidence of a possible stale head, not proof of its cause.
|
||
| - **`takeover:needs-julius` needs nothing from us.** The PR still holds one of our 3 slots until it merges or closes. | ||
| - **Keep the PR body's evidence complete** (step 3). The pipeline captures its own, but its most common stop is that it could not prove the fix, and our before and after captures and platform notes may be what lets it. | ||
| - The dashboard shows the label on the PR as GitHub has it. | ||
|
|
||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Document the temporary capacity-count limitation.
The replacement-count change is still under review and is not live. If
giveback capacitycounts the replacement as a new PR, an agent can wait at the two-PR limit even though Jackson’s rule permits another PR. Document how to check the count manually until the dashboard change is deployed.