The supplied case describes Nightshift, attributed to openslop, as a released scheduler for bounded overnight coding-agent jobs and PR reviews every two hours, with repository checks and human-controlled merging. The web snippets do not identify openslop’s project or verify those scheduling and safety rules; they surface similarly named offerings from Orbit, a LinkedIn announcement, and JudyaiLab whose relationship to this case is not established. Other snippets document overnight-agent workflows and human review practices, but do not establish that this particular scheduler makes unattended maintenance practical.
Nightshift’s claimed overnight jobs, checks and human-controlled merges repeat the operating pattern Scott already holds in The Overnight Sprint and Morning Inbox; the supplied material establishes neither a consequential new adopter nor a capability that would change his approach. No radar hit tracks Nightshift itself, but the project identity, enforced safety boundaries and practical results remain unverified, so this is currently another claimed example rather than an extension or challenge.
ip:concept.overnight-sprintip:concept.morning-inboxradar:shadok-scheduled-claude-agentsradar:moadim-git-agent-schedulerradar:e3d-pilot-sha-gated-agent-harness
queries asked of Scott's wikis
- scheduled coding-agent jobs and unattended repository maintenance
- bounded agent tasks and long-running harness orchestration
- repository checks and human approval before agent merges
- recurring automated PR review and human review bottlenecks
- cross-agent harness adapters and existing coding tools
now 0 pts/hpeak 0 pts/hcomments 0/hpeers p33momentum: steady3 platformsage 742h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
| source | object | author | score | comments |
|---|
| 🟧 hn | Show HN: Nightshift – Your code gets better while you sleepRetrieved article excerptOpen article · Retrieved 2026-09-10T18:25:19.120425+00:00 Your code gets better while you sleep. What is this? Nightshift is a helper that works at night. It wakes up your coding agent. It gives the agent one small job. The agent looks at your code and does the job. If it finds something worth fixing, it opens a pull request for you to look at. Then it moves to the next job. In the morning, you read what it did. Many nights it finds nothing. That is fine. Nothing is better than a bad change. Every two hours, it also reads any open pull requests and leaves a short review. The jobs Job What it looks for security Real holes a bad actor could use. Not "maybe someday" worries. bugs Bugs a user would notice. Only clear ones. maintainability Code that is hard to change. It makes it simpler. performance Slow spots people feel. It measures first. conventions Places that break your own written rules. Safe fixes only. conventions-followup The rule breaks that need a real change. One theme a night. architecture The one big way the code could be simpler. smoke-tests A few calm tests that check the app still opens. Usually does nothing. issues One small, clear GitHub issue. Leaves the easy ones for new people. review Reads open pull requests every two hours. Quiet unless it finds something. Every job follows the same rules. They are in jobs/_common.md . In short: Read the repo's own guide files first. They win. Read what the owner said on past pull requests. Learn their taste. Never touch a file that someone else's open pull request touches. Run the repo's checks. Fix what breaks. Open a pull request, or say why not. Never push to the main branch. Watch it Nightshift comes with a console. It shows every night and every job on one screen, and it lets you
pick up any night's conversation where the agent left off. Tell your coding agent: Open the Nightshift console. Read ~/nightshift/tui/README.md and start it. It needs Node. Press c on a job to talk to the agent about what it did. Press ? for the rest. How to start Tell your coding agent this: Get github.com/openslop/nightshift and set it up for my code at ~/code/my-app. Read SETUP.md there and follow it. That is it. Your agent will ask you a few questions, set two timers on your computer, and show you where the notes go. Works on Mac, Linux, and Windows. Works with Claude Code, Codex, Cursor, Gemini, Copilot, Hermes, OpenClaw, and most other coding agents. The full list is in nightshift.conf.example . What is inside jobs/ has one page per job, in plain English. Read them. Change them if you want. jobs/_common.md has the rules every job follows. review/SKILL.md says how the reviewer acts. nightshift.conf.example is the settings page. Your agent fills in a copy. SETUP.md is the page your agent follows to set things up. bin/ has three small scripts the timers run. You do not need to read them. tui/ is the console. tui/README.md says how to open it. Is it safe? It only ever opens pull requests. You decide what gets merged. It never touches a file that another open pull request touches. It runs your checks before it opens anything. It runs once a night. If your computer was asleep, it skips that night instead of running in the middle of your day. Everything it does is written to a log on your computer, outside your repo. Change it Turn a job off by taking its name out of the JOBS line in your settings. Change how a job thinks by editing its page in jobs/ . Keep the words plain. Add a job by adding a new page in jobs/ and putting its name in JOBS . Use a different agent by picking a different AGENT line in your settings. Turn it off Tell your agent: "Turn off Nightshift." It removes the two timers. Your code is not touched. License MIT . Free to use, copy, and change. | tcbrah | 11 | 16 |
| 🟧 echo.github ⭐ | Nightshift schedules small coding-agent jobs overnight and PR reviews every two hours; its documented rules require repository checks, avoid | openslop | — | — |
| 🟧 hn | Show HN: Knecht, a tool for building software factories for agencies | samuelreichor | 2 | 0 |
| 🟠 reddit | I'm a Claude agent that has run 100 unattended Claude Code sessions over 33 days. What held and what didn't. ClaudeAI | gaflith | 0 | 6 |
| 🟠 reddit | I gave Claude Code the night shift on my game site. Here's what it did while I slept ClaudeAI | TheRealXmo | 0 | 3 |
2026-10-10T20:08:05Z
New Reddit field report (TheRealXmo) shows Claude Code running overnight health checks, debugging, moderation, and drafting — another independent confirmation that the unattended-maintenance pattern works in practice. However, this is a direct agent-cron setup, not Nightshift, and adds no evidence on Nightshift's scheduler, safety enforcement, cross-agent reliability, or adoption. The case's distinct claim (Nightshift makes this practical across agents) remains unvalidated; pattern corroboration already exists radar-wide via sibling cases and prior field reports.
2026-10-10T15:42:12Z
evidence attached: reddit.post.1x2h7ig — Real-world overnight autonomous agent session: health checks, bot debugging, content moderation, email drafting — exactly the unattended maintenance pattern the case tracks.
2026-09-24T14:03:44Z
The 33-day, 100-session field run corroborates the case's underlying mechanism — scheduled unattended sessions sustained by exit-code checks, where note-based warnings measurably fail — but it is an agent-authored self-report received at 0 points with a 13% upvote ratio, and it validates the pattern, not Nightshift. Pattern-level support accrues across the periphery (Knecht, this field run, three sibling radar cases) while Nightshift-specific adoption and safety enforcement remain unobserved, so the case's meaning is unchanged for Scott.
2026-09-24T13:28:09Z
evidence attached: reddit.post.1wp1774 — Independent field evidence on the case's core claim: 100 unattended scheduled sessions over 33 days succeeded, with the operational finding that exit-code checks hold where note-based warnings measurably fail.
2026-09-15T09:22:38Z
Knecht adds a separate builder’s claim of sandboxed, runtime-capable repository agents, not validation of Nightshift’s scheduled maintenance or safety boundaries. This broadens the surrounding implementation landscape without changing Nightshift’s practical credibility or relevance to Scott.
2026-09-15T09:22:16Z
evidence attached: hn.story.49709714 — Independent self-hosted artifact combines triggered multi-repository coding agents, isolated execution, browser validation, and deterministic workflow steps, materially contextualizing unattended repository maintenance.
2026-09-11T06:29:48Z
The refreshed discussion raises a plausible limitation—nightly repairs need not prevent daytime agents from repeating mistakes—but supplies no observed failures, maintenance results or independent adoption. Nightshift remains a documented implementation of Scott’s existing overnight-agent pattern, with practical effectiveness and safety enforcement unvalidated.
2026-09-10T19:01:01Z
The retrieved project README establishes Nightshift’s public identity and documented workflow, but its safety rules remain instructions rather than demonstrated enforcement. No new substantive evidence establishes practical maintenance results or independent adoption; this remains an implementation of Scott’s existing overnight-agent pattern, not a reason to change his approach.
2026-09-10T18:42:52Z
grounded: known/low — Nightshift’s claimed overnight jobs, checks and human-controlled merges repeat the operating pattern Scott already holds in The Overnight Sprint and Morning Inb
2026-09-10T18:38:25Z
case created — The usable repository supplies a concrete scheduled-maintenance workflow distinct from existing session managers and multi-agent coordination cases.