[github-events] ingest status webhook (commit-status / external CI) [#2321]

[lsm]

Subscribes to the GitHub status webhook (the commit-status API used by external/legacy CI — Jenkins, Travis, custom) and re-expresses it as pull_request/<id>.status_<state>, surfacing pending as well as failure/error/success.

The payload addresses a commit SHA, not a PR, so handleStatusWebhook resolves the SHA to the open PR(s) whose head it is via /commits/{sha}/pulls (filtering out PRs that merely merged the commit) before publishing. Empty or failed resolution drops the event. status is added to WEBHOOK_EVENTS/REQUIRED_WEBHOOK_EVENTS so auto-managed hooks subscribe and re-register for it.

Closes #2321.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[greptile-apps]

Greptile Summary

This PR subscribes to the GitHub status webhook (the commit-status API used by external/legacy CI systems like Jenkins and Travis) and re-expresses each delivery as pull_request/<id>.status_<state> events. Because the status payload addresses a commit SHA rather than a PR, a new handleStatusWebhook method resolves the SHA to open PRs via /commits/{sha}/pulls (paginated, up to 5 pages) before normalizing and publishing. The resolution only fires when at least one watching space is enabled, matching the quota-conservation intent of the non-status paths.

• New webhook handler: handleStatusWebhook guards enabled/space checks before the SHA→PR API call, filters out merged-commit false positives by comparing head.sha, and emits one status event per (watched-space × PR) pair.
• Normalizer addition: normalizeGitHubStatus structures the payload with a per-PR deduplication key (status:<id|sha:context>:<state>:<prNumber>) and uses context as fallback in the identity when id is absent.
• Downstream wiring: SQL tone/title rules, isReactivePrCheckFailure regex, and formatExternalEventEssence all extended to handle status_failure / status_error as CI failures, with status_pending / status_success treated as neutral PR updates.

Confidence Score: 5/5

Safe to merge — the status webhook ingestion is well-guarded, the two previously identified issues have been addressed, and the new code paths are thoroughly tested.

The handler correctly gates the SHA→PR API call behind enabled-space checks, deduplication keys are properly scoped per (CI-system × PR), and all four commit-status states are routed consistently through the existing tone/reactivation logic. Test coverage spans the full branching surface of the new code.

Files Needing Attention: No files require special attention.

Important Files Changed

Filename	Overview
packages/daemon/src/lib/external-events/github/github-event-extension.ts	Adds handleStatusWebhook and resolvePullRequestNumbersForCommit — the core of the new feature. Enabled/space checks are correctly placed before the SHA→PR API call, pagination is bounded, and merged-commit false positives are filtered by comparing head.sha.
packages/daemon/src/lib/external-events/github/github-normalizer.ts	Adds normalizeGitHubStatus with a per-PR deduplication key scoped by id (or sha:context fallback), and routes status through mapEventType as status_. repoFromPayload is exported for use in the extension.
packages/daemon/src/lib/space/runtime/space-runtime.ts	Expands isReactivePrCheckFailure regex to also match .status_failure and .status_error topics, correctly excluding status_pending and status_success from task reactivation.
packages/daemon/src/lib/rpc-handlers/live-query-handlers.ts	SQL tone/title CASE expressions updated to treat .status_failure and .status_error topics as danger / CI check failed, matching the semantics of .check_failed. status_pending correctly falls through to neutral PR update.
packages/daemon/src/lib/external-events/event-essence.ts	Adds the status event type to formatExternalEventEssence, copying the right fields. Clean, consistent with adjacent branches.
packages/daemon/tests/unit/2-handlers/github/github-event-extension.test.ts	New tests cover: failure/pending resolution, empty-result drop, merged-commit filter, unwatched-repo 404, and the disabled-target quota guard.
packages/daemon/tests/unit/2-handlers/github/github-normalizer.test.ts	Tests all four states, per-PR dedupe scoping, id-absent CI collision fix, commit.sha fallback, and null returns for missing fields.
packages/daemon/tests/unit/2-handlers/rpc-handlers/live-query-handlers.test.ts	Adds a test verifying that status_failure/status_error render as danger/CI check failed and status_pending renders neutral.
packages/daemon/tests/unit/5-space/runtime/space-runtime-external-events.test.ts	Two new tests verify that status_failure reactivates a done task and that status_pending/success do not.

Sequence Diagram

sequenceDiagram
    participant GH as GitHub
    participant WH as Webhook Handler
    participant EXT as handleStatusWebhook
    participant DB as WatchedRepo DB
    participant API as GitHub API
    participant PUB as EventPublisher

    GH->>WH: POST /webhook/github/space
    WH->>WH: verify signature to signatureMatchedRepos
    WH->>EXT: handleStatusWebhook(deliveryId, payload, matchedRepos)

    EXT->>EXT: extract repo + sha from payload
    alt repo or sha missing
        EXT-->>GH: 202 Event ignored
    end

    EXT->>EXT: filter matchedRepos to validForRepo
    alt validForRepo empty
        EXT-->>GH: 404 Repository not watched
    end

    loop each watched in validForRepo
        EXT->>DB: check watched.enabled
        EXT->>DB: getSpaceConfig(spaceId)
        note over EXT: collect into targets[]
    end

    alt targets empty
        EXT-->>GH: 200 spaces:0
    end

    EXT->>API: GET /commits/sha/pulls
    API-->>EXT: PR list filtered by head.sha

    alt prNumbers empty
        EXT-->>GH: 202 no_pull_request
    end

    loop each watched in targets
        loop each prNumber
            EXT->>EXT: normalizeGitHubStatus()
            EXT->>PUB: publishEvent(spaceId, normalized)
        end
        EXT->>DB: markWebhookReceived(watched.id)
    end

    EXT-->>GH: 200 spaces:N

Loading

_{Reviews (3): Last reviewed commit: "fix(space): wire external-CI status fail..." | Re-trigger Greptile}

[lsm]

🤖 Review by glm-5.1 (NeoKai)

Model: glm-5.1 | Client: NeoKai | Provider: z.ai

Recommendation: REQUEST_CHANGES

The ingestion itself is clean and well-tested — webhook subscription, SHA→PR resolution with correct head-sha filtering (excludes merged-commit false positives), PR-scoped dedupe, all four states including pending, and a matching essence projection. The two Greptile P2s are already resolved in 68670f5a (enabled-guard short-circuit before the resolution GET; context-aware dedupe fallback) with covering tests. I re-ran the two test files (195 pass) and typecheck/lint are clean.

Two integration findings remain — both about making external/legacy CI failures as actionable as native check_run failures:

P1 — status_failure / status_error don't reactivate a done task

isReactivePrCheckFailure (packages/daemon/src/lib/space/runtime/space-runtime.ts:779) hard-matches .check_failed only:

/^github\/[^/]+\/[^/]+\/pull_request\/[^/.]+\.check_failed$/i.test(event.topic)

A done PR task is reactivated on a native check_run failure, but a status_failure/status_error from Jenkins/Travis/custom CI is dropped as target_task_terminal in prepareExternalEventTask (the reactive path is the only action-sensitive branch). Since surfacing external CI is the whole point of #2321, a legacy-CI failure silently doing nothing while a GitHub-Actions failure reopens the task is a confusing gap — the events are ingested and delivered to in-progress agents, but they can't reopen a completed one.

Suggested fix — widen the predicate to also match the failure states (not pending/success):

/^github\/[^/]+\/[^/]+\/pull_request\/[^/.]+(?:\.check_failed|\.status_(?:failure|error))$/i

If you intend to defer this, please open a tracked follow-up and call it out in the PR description — but my read is that it belongs in the PR that introduces status_failure.

P2 — Activity feed renders external CI failures as neutral "PR update"

packages/daemon/src/lib/rpc-handlers/live-query-handlers.ts:1347-1358 special-cases only %.check_failed as danger / "CI check failed". A status_failure/status_error falls through to WHEN ee.topic LIKE '%pull_request%' THEN 'PR update' → neutral tone. External CI failures should get the same treatment so the feed surfaces them as failures, e.g.:

WHEN ee.topic LIKE '%.check_failed'
  OR ee.topic LIKE '%.status_failure'
  OR ee.topic LIKE '%.status_error' THEN 'danger'

(and the matching 'CI check failed' title branch).

---

Reviewed against working tree @ 68670f5a (both commits). Tests: 195 pass / 0 fail across the two changed test files; tsc + oxlint clean.

[lsm]

🤖 Review by glm-5.1 (NeoKai)

Model: glm-5.1 | Client: NeoKai | Provider: z.ai

Recommendation: APPROVE

Both findings from my prior review are resolved in 8ff37547a, re-verified fresh this round (zero new findings):

• P1 — isReactivePrCheckFailure (packages/daemon/src/lib/space/runtime/space-runtime.ts:779) now matches .check_failed | .status_failure | .status_error via [^/.]+\.(?:check_failed|status_(?:failure|error))$, so an external-CI failure reopens a done PR task like a native check_run; pending/success stay excluded. Reactivation test (status_failure) + negative test (status_pending stays terminal) added.
• P2 — activity-feed SQL (packages/daemon/src/lib/rpc-handlers/live-query-handlers.ts) now classifies .status_failure/.status_error as danger / "CI check failed", matching check_failed; pending renders neutral. Tests added.

Verified: 294 pass / 0 fail (live-query-handlers + space-runtime-external-events), 195 pass / 0 fail (github ingestion tests); tsc, oxlint, knip all clean; 0 unresolved review threads. The ingestion logic and the earlier Greptile fixes (68670f5a) remain intact.

Nice work — external/legacy CI is now first-class alongside native check_run across ingestion, reactivation, and the activity feed.