| name | create-contributor-prs |
| description | For an external contributor's open PRs to brave/brave-core, ensure the `sync-and-rebase-pr-from-fork.yml` workflow has been dispatched for each PR's latest commit SHA — so a `contributor-*` draft PR exists inside brave/brave-core and CI can run. Invoke when the user says something like "create contributor PRs for <user>", "sync contributor PRs <list>", "run the rebase-from-fork workflow for these PRs", or passes a GitHub username / list of PR URLs/numbers. |
create-contributor-prs
External contributors fork brave/brave-core and open PRs from their fork. Most
of Brave's CI doesn't run on fork PRs for security reasons. The
sync-and-rebase-pr-from-fork.yml workflow exists to create an in-org mirror
branch (contributor-<owner>-<headRefName>) and a draft PR in brave/brave-core
that re-triggers full CI on the rebased commit.
This skill dispatches that workflow for each open PR by a contributor (or each
PR in an explicit list), but only when a dispatch hasn't already happened for
the PR's current HEAD commit.
Inputs
The user invokes this skill with one of:
- A GitHub username — e.g.
create-contributor-prs jharris-tech. Resolve
to all open PRs they authored against brave/brave-core.
- A list of PRs — any mix of:
- Full URL:
https://github.com/brave/brave-core/pull/36402
- Short form:
#36402 or 36402
- Cross-repo form:
brave/brave-core#36402
If the user gives both a username and PR list, treat them as additional filters
(union of both).
If no input is given, ask for one. Don't proceed without explicit scope.
Repo
Always brave/brave-core unless the user explicitly names a different repo.
Pass --repo brave/brave-core to every gh call. Don't assume the cwd matters.
Workflow facts (don't re-derive)
.github/workflows/sync-and-rebase-pr-from-fork.yml inputs:
PR_NUMBER (optional) — number of the contributor PR
COMMIT_HASH (required, 40-char hex) — the contributor PR's HEAD commit SHA
The workflow:
- Requires the PR to be cross-repo (
isCrossRepository: true). It exits 1
otherwise — so filter non-cross-repo PRs out client-side, don't dispatch them.
- Checks out the given
COMMIT_HASH from the contributor's fork, rebases onto
the PR's base branch, force-pushes to
refs/heads/contributor-<headRepositoryOwner>-<headRefName> inside
brave/brave-core.
- Creates a draft PR titled
CI run for contributor PR #<PR_NUMBER> if one
doesn't already exist for that contributor branch.
GitHub does NOT expose workflow_dispatch inputs via the runs API. To find out
which COMMIT_HASH a past run used, you must read the run's log.
Procedure
Step 1 — Resolve PRs
For a username:
gh pr list --repo brave/brave-core --author <USER> --state open \
--json number,title,url,headRefOid,headRepositoryOwner,headRepository,isCrossRepository,headRefName \
--limit 100
For each user-supplied PR number N:
gh pr view N --repo brave/brave-core \
--json number,title,url,headRefOid,headRepositoryOwner,headRepository,isCrossRepository,headRefName,state,author
Combine into one list. Drop any PR where isCrossRepository == false (the
workflow rejects same-repo PRs) and report them as "skipped: not cross-repo".
Drop any PR not in OPEN state and report it as "skipped: not open".
Step 2 — Check whether the workflow already ran for each PR's current HEAD
For each PR, compare its headRefOid against the COMMIT_HASH inputs of recent
workflow runs.
Pull a window of recent runs once, up front (don't refetch per-PR):
gh run list --repo brave/brave-core \
--workflow=sync-and-rebase-pr-from-fork.yml \
--limit 100 \
--json databaseId,status,conclusion,createdAt,event
For each candidate run (filter to event == "workflow_dispatch"), the
COMMIT_HASH input appears in the job log on a line like
COMMIT_HASH: <40-hex>. Fetching the full log per run is slow, so batch-fetch
only as many as needed: iterate runs newest-first and stop early once every
target PR's HEAD SHA is either matched or determined absent in the window.
Per-run log fetch + extract:
gh run view <RUN_ID> --repo brave/brave-core --log 2>/dev/null \
| grep -oE 'COMMIT_HASH: [0-9a-f]{40}' \
| head -1 \
| awk '{print $2}'
Build a set dispatched_shas. Also record per-SHA whether the run is
in_progress / queued / completed (success or failure) — useful for
reporting.
If the run window of 100 doesn't cover far enough back (e.g. the contributor's
HEAD commit is older than the oldest run in the window), expand the window with
--limit 200 etc., or accept that you may re-dispatch. The workflow is
idempotent (force-push + skip-if-exists draft PR), so a duplicate dispatch is
wasteful CI but not destructive. Mention this tradeoff if you bail on the
lookback early.
Step 3 — Dispatch the workflow where needed
For each PR whose headRefOid is NOT in dispatched_shas:
gh workflow run sync-and-rebase-pr-from-fork.yml \
--repo brave/brave-core \
-f PR_NUMBER=<N> \
-f COMMIT_HASH=<headRefOid>
gh workflow run returns nothing useful on success and exits non-zero on
failure. Capture stderr.
Do NOT pass --ref — let it default to the workflow's default branch (master).
The workflow operates on the input commit hash, not the ref.
Step 4 — Report
Print a compact table per PR:
PR SHA Status
36402 3807075d already dispatched (in_progress, run 26407241424)
36510 abc1234d dispatched now
36588 9988aabb skipped: not cross-repo
36601 77665544 skipped: closed
End with a one-line summary: Dispatched N, already-running M, skipped K.
If any dispatch failed, list the PR + the stderr from gh workflow run.
Confirmation before dispatching
Each dispatch consumes CI minutes and creates/updates a draft PR in the org.
Before triggering a batch dispatch, show the list of PRs you're about to
dispatch for and ask the user to confirm — unless they explicitly said "just do
it" / "no confirmation" / passed something like --yes. A single-PR dispatch
can proceed without re-confirming if the user clearly named that one PR.
Things to avoid
- Don't dispatch for non-cross-repo PRs — the workflow exits 1.
- Don't dispatch for closed/merged PRs.
- Don't use
display_title or head_sha of the workflow run to infer the
COMMIT_HASH input. head_sha is the SHA of master at dispatch time, not the
input. Only the log has the real input.
- Don't fetch every run's log up front — stop early when all target SHAs are
resolved.
- Don't comment on or modify the contributor's PR. This skill only triggers a
workflow.