Engineering Outer Loop Dot
Your ChatGPT dot keeps PRs, CI and Linear tidy and brings you only the decisions.
作者 GetKits
看看前后差别
Support DM to a fix waiting for your yes
之前
Slack DM from Sam (support), 2:12 PM: "Lumen says CSV exports from the Orders page are missing the last row. Every time, any date range. Ticket #8812, I filed WEB-512."
之后
Reproduced: Yes. A test exporting 3 orders produced 2 data rows. Suspected cause: src/orders/export/toCsv.ts:41 loops with i < rows.length - 1, left over from #803. Status: Branch fix/WEB-512-csv-last-row prepared with regression tests, not pushed. pnpm test src/orders/export: 18 passed. Decision needed: Push the branch and open a draft PR?
9:15 check-in with two decisions
之前
Scheduled check-in, Wed 2026-09-30, 9:15 AM Central. GitHub: platform main green since yesterday 08:51 (CI #5126, fixed by #1952). Your PR platform#1955 "refactor(jobs): extract the shared scheduler" has 2 approvals but now conflicts in src/jobs/index.ts after #1952. platform#1931 (Priya, size S) has waited 76 h on Marco, who is out until Thursday. Linear: PLAT-340 "Webhook signature mismatch after rotation" set to Urgent by support at 08:40, now 4 customers. Open PLAT-311: sender keeps the old secret for 15 minutes after rotation. Dana owns webhooks.
之后
REPO TRIAGE Wed 2026-09-30 9:15 AM CT northwind/platform, northwind/web Main: platform green since Tue 08:51 (CI #5126). web green. Needs review (oldest first, SLA 24 h): - platform#1931 fix(billing): round line items before tax. Priya, waiting 76 h, S, reviewer Marco (out until Thursday). Over SLA Blocked: - platform#1955 refactor(jobs): extract the shared scheduler (yours). 2 approvals, merge conflict in src/jobs/index.ts after #1952 Top 3 today: 1. Decide who takes PLAT-340 (below) 2. Rebase #1955 3. Reassign the review of #1931 (below) Sources: GitHub northwind/platform, northwind/web; Linear PLAT, WEB; #eng-alerts. Nothing missing. DECISIONS NEEDED 1. Decision: Who takes PLAT-340 (webhook signature mismatch, Urgent since 08:40, 4 customers)? Options: A) Dana, who owns webhooks. B) Wait for the on-call rotation to pick it up. Recommendation: A. Four customers have no workaround, and PLAT-311 (old secret used for 15 minutes after rotation) is the likely cause, which Dana knows. By when: 10:00 AM today Links: PLAT-340, PLAT-311 2. Decision: Reassign the review of platform#1931? Options: A) Ask Ana to review today. B) Wait for Marco on Thursday. Recommendation: A. It is size S, already 52 h over SLA, and Marco is out until Thursday. By when: noon today Links: platform#1931
A bug it cannot reproduce, stopped after two tries
之前
Slack DM from Sam (support), Tue 2026-10-13, 10:05 AM: "Acme says applying a discount code at checkout sometimes does nothing. No error, the total just stays the same. Maybe 1 try in 10, and trying again usually works. Ticket #8937, I filed WEB-544." You, 10:07 AM: "Investigate and prepare a fix if you find it." Your dot has a Codex cloud environment for northwind/web.
之后
BUG INVESTIGATION Report: Applying a discount code at checkout sometimes leaves the total unchanged, with no error, about 1 try in 10 (Acme, ticket #8937, WEB-544). Reproduced: No, after two attempts in the Codex cloud environment for northwind/web. 1. Ran checkout.spec.ts "applies discount code" 25 times on current main: 25 passed, the total changed every time. 2. Throttled the network to slow 3G and clicked Apply the moment it appeared, 10 times: the discount applied all 10 times. Suspected cause: Not confirmed. One lead: the same test, checkout.spec.ts:88, failed on main 3 times in the last 30 days (runs #2190, #2197, #2205), each a timeout waiting for the Apply button and green on rerun. That points at timing around the discount form, but neither attempt reproduced it. Proposed fix: None. No code changes without a reproduction. Risk: Not assessed, no change proposed. Test plan: Once one failing try reproduces, write a test that fails before the fix and passes after, then run pnpm test src/checkout. Status: No branch, no PR, nothing pushed. Stopped after two attempts, as instructed (2 Codex tasks). Decision needed: Send the question below to Sam and add it as a comment on WEB-544? Draft for Sam (not sent): "Can you get one failed try from Acme: the time with time zone, the browser and version, the exact code, and whether it worked after a page refresh? A screen recording or a HAR file from the browser's network tab is the fastest way for us to find it." ISSUE TRIAGE Issue: WEB-544 Discount code sometimes does nothing at checkout (Acme) Type: bug Priority: Medium. Trying again usually works, so there is a workaround. Raise to High if a second customer reports it. Team: WEB Labels: bug, checkout Duplicate of: none found in Linear WEB or northwind/web issues Missing: one failed try with time, browser and code Proposed action: set Medium and add the question as a comment, when you say so.
Release day: notes plus the go or no-go
之前
Thursday 1:00 PM check-in, release day. Merged into northwind/platform since v4.7.0: #1931 fix(billing): round line items before tax #1941 refactor(api): rename nextPage to nextCursor #1944 feat(api): cursor pagination for /v2/orders #1950 chore: move cron config to jobs/ #1952 fix(sdk): use nextCursor in orders client (fixed main, broken by #1944) #1961 feat(webhooks): retry failed deliveries with backoff (behind WEBHOOK_RETRY_ENABLED, default off) #1963 chore(deps): bump pg from 8.11.3 to 8.12.0 #1964 test: webhook retry fixtures #1966 feat(webhooks)!: drop the X-Northwind-Signature-SHA1 header (body: use X-Northwind-Signature-256; Fernway, Acme and Lumen still verify SHA-1) #1968 fix(webhooks): use the new secret immediately after rotation (PLAT-340) #1970 fix (empty body) Release cut at 3:00 PM.
之后
RELEASE NOTES Version: v5.0.0 suggested (breaking change present). Range: v4.7.0...main, 11 PRs Breaking changes - Webhooks no longer include the X-Northwind-Signature-SHA1 header. Receivers that verify SHA-1 signatures must switch to X-Northwind-Signature-256 before upgrading. (#1966) New - List /v2/orders with cursor pagination. (#1944) Fixed - Invoice totals now match checkout when line items have fractional cents. (#1931) - Webhooks verify immediately after you rotate your signing secret, instead of failing for up to 15 minutes. (#1968) Internal (5 changes): nextCursor rename, cron config move, SDK fix within this release, pg 8.12.0, retry fixtures (#1941, #1950, #1952, #1963, #1964) Needs confirmation - #1961 webhook retries: behind WEBHOOK_RETRY_ENABLED, default off. Is it on in production for this release? - #1970 "fix": empty description. What changed for users? Check: 11 PRs in, 11 accounted for (1 breaking, 1 new, 2 fixed, 5 internal, 2 to confirm) DECISIONS NEEDED 1. Decision: Ship the SHA-1 header removal (#1966) today? Options: A) Ship v5.0.0 with it; Fernway, Acme and Lumen stop verifying until they switch. B) Revert #1966 for this release and ship v4.8.0; remove SHA-1 once the three accounts confirm. Recommendation: B. Three named accounts still verify SHA-1, and Lumen already hit PLAT-340 this week. By when: 3:00 PM release cut Links: #1966, PLAT-340 2. Decision: Announce webhook retries (#1961)? Options: A) Turn the flag on in production and list retries under New. B) Keep it off and leave it out of the notes. Recommendation: B until staging has run 24 hours with the flag on, as the PR's rollout plan says. By when: 3:00 PM Links: #1961
包含内容
- engineering-outer-loop-dot-setup.md
关于这个 Kit
Give your ChatGPT dot the ongoing job of keeping your team's outer loop clean. Each weekday at 9:15 it checks your repos, Linear and your alerts channel and updates the triage. It keeps PR descriptions and a running release notes draft current, and investigates reported bugs and prepares a fix for code review. Merge, release and priority decisions come to you as a short list with options and a recommendation. It asks before it comments, pushes or opens a PR, and it never merges or deploys. Setup takes about 15 minutes.
What you get
- A responsibility block that names the sources (GitHub, Linear, Slack), what to keep current, which changes deserve your attention right away, and where updates go: routine progress in ChatGPT, decisions in Slack.
- Working instructions with six fixed output blocks: REPO TRIAGE, PR DESCRIPTION, RELEASE NOTES, ISSUE TRIAGE, BUG INVESTIGATION and DECISIONS NEEDED. They match the Engineering Outer Loop Skill and Bot field for field.
- A bug investigation method: reproduce first, cite the file and line, make the smallest fix with a test that fails before and passes after, then a ready PR description. It uses a Codex cloud environment you have created or your connected computer, and stops after two failed reproduction attempts.
- Eleven custom rules set in Settings: reading repos happens without asking, Linear triage when you say so, comments, pushes, PRs, Slack posts and reruns ask first, and merges or deploys are handed off to you.
- A check-in schedule: weekday triage, a release-day go or no-go, and a monthly hygiene review, with a time zone and an end date.
- A first-week review plan, including a calibration run on a bug your team already fixed.
- Everyday prompts and a troubleshooting list covering Slack monitoring that never starts, tests that cannot run in the cloud environment, wrong check-in times, and how to stop work cleanly.
Example of what reaches you in Slack: "DECISIONS NEEDED 1. Ship the SHA-1 header removal today? Recommendation: hold it for this release; three named accounts still verify SHA-1. By 3:00 PM."
Who it is for
- Tech leads and engineering managers who want one place to see what needs them, without reading every notification.
- Engineers on a ChatGPT Pro or Business Premium plan who already use their dot and want it on the engineering side of their work.
- Teams that use GitHub and Linear and talk in Slack.
- Leads who want bug reports from support turned into a reproduced cause and a reviewable fix before they reach the sprint board.
What it will not do
- Merge, deploy, close issues or change repo settings. Those are handed off to you.
- Comment, push a branch, open a PR, post in a channel or rerun CI without asking first.
- Set Linear priorities on its own. It proposes them and applies them when you say so.
- Claim a test passed that it did not see pass, or invent issue keys and owners. Unknown facts are marked
[confirm]. - Replace code review. Every fix it prepares goes through your normal review.
评价
还没有评价。
更多工程类 Kit
Engineering Outer Loop Kit
Run your team's outer loop: daily triage, CI verdicts, release notes and postmortems.
Engineering Outer Loop Bot
A Grok teammate that triages your repos at 9:00 and drafts release notes. It never merges.
Engineering Outer Loop Skill
Turn diffs, red CI logs and merged PRs into review-ready write-ups with the risk named.