Nightshift Coding Shifts

Keep coding agents on a bounded task list and return to progress you can inspect.

Stars
★ 551
Last updated
1d ago
License
MIT
Primary language
Shell

At a glance

How it runs
Agent plugin / skill
Works with
Universal · cross-platformCodex · Claude Code
Cost
Free software; you pay for model usage
Setup effort
Low · running in minutes
You'll need
Claude CodeOpenAI CodexCursorBashPowerShellShell / CLILocal filesystem
Typical use
Developers who want Claude Code, Codex, or Cursor to work through several hours of bounded coding tasks can use a finite list and deadline.
Not a fit if
  • Teams that require human approval at every stopping point
  • Users who cannot allow an agent to access project files
Website
nightshift.orwamahmoud.com ↗ set on GitHub
Maintenance
Commits in the last 30 dayslast commit 2026-11-19latest release v0.37.2 (2026-11-19)+19.6 ★/day over 7 days
Source review
62/100 · Some gaps

What does this agent do, and when should you use it?

Nightshift is a plugin for Claude Code, OpenAI Codex, and Cursor that gives long coding runs a persistent work contract. Users can shape tasks in a plan room, write a punch list, or choose a ready-made shift such as Hunt or Quality, then set an objective or time limit. Work proceeds item by item, with verification requests, evidence records, time or token budgets, and a clock-out gate that can hold unfinished work. Progress, decisions, and receipts live in project files under .nightshift; a watchman can recover sessions when it sees supported failure evidence. Nightshift runs on macOS, Linux, and Windows without a separate service or API key, and users review the results before deciding whether to push.

After setup in a project, configure checks and defaults with natural language or Claude Code's /nightshift:setup. Plan work in the plan room or write items under ## Items in .nightshift/punch-list.md. Start execution by asking Codex “Start the Nightshift shift.” or running /nightshift:start in Claude Code. Each item can specify Verify checks, a Commit message, evidence kinds, and soft or hard time or token limits; execution records state, changed files, attempts, and verification in receipts. Use /nightshift:status to inspect progress and /nightshift:stop to stop while leaving unfinished items open. Review the morning receipt and item receipts, then archive and continue. Git workspaces can set shift.isolation to worktree; persistent folders without Git can use artifact receipts.

  1. Developers who want Claude Code, Codex, or Cursor to work through several hours of bounded coding tasks can use a finite list and deadline.
  2. Project maintainers with a backlog can use Hunt to compose work from catalogs covering defects, dependency upgrades, documentation, or accessibility.
  3. Teams that want to review task breakdowns before coding can use the plan room's pre-mortem, edge-case pass, and ordering check.
  4. Developers whose runs span context compaction or session failures can use the persistent punch list, parked decisions, and watchman recovery.
  5. Maintainers who need to inspect agent output item by item can review morning and item receipts for evidence, verification, file maps, and available usage measurements.

How do you install or deploy this agent?

In Claude Code, add the marketplace and install the plugin:

/plugin marketplace add orwa-mahmoud/nightshift
/plugin install nightshift

In Codex, add the marketplace and install:

codex plugin marketplace add orwa-mahmoud/nightshift
codex plugin add nightshift@nightshift

In Cursor, open Customize → Add → From GitHub Repository, enter the repository URL, choose Import, then install Nightshift from the imported marketplace. Open the Git repository or persistent local folder the agent should change. macOS and Linux use the bundled Bash runtime; native Windows uses bundled PowerShell. The repository says no separate service or API key is needed; model usage costs depend on the host.

How do you use this agent?

In Claude Code, set up and start a five-hour improvement shift:

/nightshift:setup
Improve this project for 5 hours.
/nightshift:start

Alternatively, run /nightshift:plan to design the work, then /nightshift:plan-exit to leave the plan room. Check progress with /nightshift:status, diagnose with /nightshift:doctor, and stop with /nightshift:stop. In Codex, use natural-language requests to set up, plan, start, and inspect status; items can also be written under ## Items in .nightshift/punch-list.md. Choose permissions and verification before starting. Afterward, review the morning receipt and item receipts, then decide whether to push commits.

What are this agent's strengths and limitations?

Pros
  • The punch list and progress persist on disk for long runs that cross context compaction.
  • Per-item verification, evidence requirements, time or token budgets, and a clock-out gate constrain execution.
  • Morning and item receipts record deliverables, attempts, verification, changed files, and available usage data.
  • Supports Claude Code, Codex, and Cursor on macOS, Linux, and Windows.
Limitations
  • It runs through supported coding hosts and their shell environments; it is not a standalone coding interface.
  • Unattended runs cannot approve host prompts, so permissions and checks must be chosen in advance.
  • Recovery depends on host-provided failure signals and has host-specific limits.
  • Token and duration measurements vary by host, and the README says receipts do not independently prove quality.

How does this agent compare with similar options?

Compared with asking a coding agent to keep going in chat, Nightshift adds a persistent punch list, clock-out gate, failure recovery, and per-item receipts. The README describes the ChatGPT Plugin Directory edition as skills-only for attended sessions, without the clock-out gate, recovery, or usage accounting.

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
Nightshift Coding Shifts This agent 62 · Some gaps Agent plugin / skillFree + model costs ★ 551 1d ago Shell Codex · Claude Code
PandaCode 67 · Some gaps — ★ 1.3k today TypeScript Codex · Claude Code · Claude.ai
Multica — Multi-Agent Task Collaboration Platform 61 · Some gaps Web appFreemium ★ 57k today Go ChatGPT · Codex · Claude Code
Concord MCP 75 · Good — ★ 1.4k 1d ago TypeScript Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Some gaps
62/ 100 5-point scale 3.1 / 5
This review covers the source as of Nov 11, 2026. The project has shipped 6 stable release(s) since, which have not been re-reviewed. The score and verdict describe the reviewed version only. Recent releases ↓
Trust 16/29
Reliability 8/14
Adaptability 14/18
Convention 13/18
Effectiveness 7/13
Verifiability 4/8
Why each dimension lost points
Trust16 / 29 · 2.8/5

least_privilege: README and customize/doctor skills expose toolDeny, protectedDirs, forbiddenCommands and elevation categories, but defaults allow, so configurable rather than least-privilege by default — deducted. user_confirmation: install skill installs only on hosts the user agrees to; customize refuses while armed and apply-profile needs --apply — good. data_flow_transparency: SECURITY.md states no telemetry and no phone-home, and explains notifyCommand and the launched agent's network behavior — transparent. sensitive_data_handling: SECURITY.md asks reports to exclude credentials/prompts/repo content, but no concrete redaction mechanism for logs or secrets is shown — deducted. dependency_security: CI pins action commit SHAs, but runtime deps (jq/sed fallback, bats, kcov, docker) are unpinned with no SBOM — deducted. external_effects: commits stay local unless the contract authorizes, STOP wins, worktree isolation optional — clear boundaries. rollback: migrate-state preview/--apply, refuses overwrite, writes state-version last, archive copies then reads back — thorough. source_attribution: author and repo links are clear, but publisher is unverified and no third-party corroboration — deducted.

Reliability8 / 14 · 2.9/5

self_consistency: README, install, troubleshooting and skills agree on state-version, work-mode and STOP semantics — no contradiction found. dependency_availability: depends on host CLIs (claude/codex/cursor), optional jq, bats/kcov/docker for tests, with no version constraints or offline notes — deducted. failure_messages: doctor classifies Facts/Warnings/Actions and troubleshooting gives concrete repair commands with fail-closed semantics — good.

Adaptability14 / 18 · 3.9/5

audience_and_scenarios: covers Claude Code, Codex, Cursor, ChatGPT, skills.sh, WSL, remote SSH and native Windows — broad. capability_boundaries: skills-only edition lacks gate/recovery/usage and artifact mode skips several hunts; boundaries stated but scattered. trigger_precision: skills carry name/description and internal flags with a clear command table, but triggers rely on natural language — moderate. environment_fit: per-platform runtimes (Bash/PowerShell) and WSL constraints documented — good.

Convention13 / 18 · 3.6/5

information_architecture: README has TOC, docs index and vocabulary — excellent. install_notes: per-host install steps are detailed with update and removal hints. naming_stability: skill names and commands are stable, but version numbers appear only sporadically (0.25.3) and naming/version coupling is unclear. examples_and_faq: real-run examples and a punch-list example exist, but no standalone FAQ. known_limitations: README and troubleshooting list recovery limits and skills-only gaps, but no consolidated limitations list. license: full MIT text matches metadata. versioning_changelog: CI references Release Please and CHANGELOG, but no CHANGELOG content or versioning policy is in the packet — deducted. maintenance_responsibility: single author, no maintainer list or SLA, publisher unverified — deducted.

Effectiveness7 / 13 · 2.7/5

output_usability: outputs are receipts, morning receipt and archive — structured and reviewable. marginal_value: addresses long-run drop-off and premature clock-out, but depends on host hook ecosystems and alternatives exist. cost_benefit: requires plugin install, permission setup and understanding multi-file state; learning/config cost is high and benefit needs a real run to confirm — deducted.

Verifiability4 / 8 · 2.5/5

claim_traceability: README claims real runs and a 67-commit PR but only links to example docs with no independently checkable evidence — deducted. cross_source_corroboration: CI workflows and test scripts exist, but no test results or coverage data are shown — limited. fact_inference_separation: docs distinguish that a tick is a completion claim not proof of quality, and doctor separates Facts/Warnings — good.

Risks and how to mitigate them
  • Publisher identity is unverified; review hooks and runtime scripts before installing.
  • Default permission knobs mostly allow; tighten toolDeny, protectedDirs and elevation before unattended runs.
  • notifyCommand is unrestricted owner-provided shell that may access the network — configure carefully.
  • Runtime relies on jq/sed fallback and host CLIs with no version pinning or SBOM.
  • Real-run and PR claims in README lack independently checkable evidence; treat as author self-report.
Evidence confidence: Low Reviewed Nov 11, 2026 Reviewed revision 8eb6fc1ef0ea New commits since this review; the score may not cover them
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/nightshift.svg)](https://followagents.com/en/agents/nightshift)
<a href="https://followagents.com/en/agents/nightshift"><img src="https://followagents.com/badge/nightshift.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v0.37.2
    Release notes excerpt
    [0.37.2](https://github.com/orwa-mahmoud/nightshift/compare/v0.37.1...v0.37.2) (2026-11-19) Bug Fixes * record the saved verification profile and tooling policy at Start ([aa665cf](https://github.com/orwa-mahmoud/nightshift/commit/aa665cf2b5c60380ffc04d7cd8e7ff30a729d96c)) * refuse an unreadable hook payload with the true reason on every runtime ([1d3f711](https://github.com/orwa-mahmoud/nightshift/commit/1d3f7119cbd0c6e09ee91c61c41ee648d52917e8))
    View on GitHub ↗
  2. v0.37.1
    Release notes excerpt
    [0.37.1](https://github.com/orwa-mahmoud/nightshift/compare/v0.37.0...v0.37.1) (2026-11-19) Bug Fixes * capture UI evidence at the viewport and 1x scale ([94fe7cf](https://github.com/orwa-mahmoud/nightshift/commit/94fe7cf8e2a11643fca7a496f151dcf9ab2cb8c0))
    View on GitHub ↗
  3. v0.37.0
    Release notes excerpt
    [0.37.0](https://github.com/orwa-mahmoud/nightshift/compare/v0.36.0...v0.37.0) (2026-11-18) Features * add a changelog tag drift shift that reconciles notes with the last tag ([4c1fdb5](https://github.com/orwa-mahmoud/nightshift/commit/4c1fdb5faa1acb03a17849979b112a5715b5ec1e)), closes [#21](https://github.com/orwa-mahmoud/nightshift/issues/21) * add a changelog tag drift shift that reconciles notes with the last tag ([#236](https://github.com/orwa-mahmoud/nightshift/issues/236)) ([5b4dc4f](https://github.com/orwa-mahmoud/nightshift/commit/5b4dc4fd071d092549339c55d4919e6f11d9e2a7))
    View on GitHub ↗

FAQ

Does Nightshift need a separate API key or service?
The README says no separate service or API key is needed; model costs depend on how Claude Code, Codex, or Cursor is used.
Can it approve permission prompts while unattended?
No. The README says unattended runs cannot approve prompts, so choose permissions and checks beforehand.
Will it always recover a failed session automatically?
No guarantee is documented. The watchman requires positive failure evidence, and recovery depends on each host's supported signals and limits.
Does Nightshift push code automatically?
Commits stay local by default. Review the results and push when you decide; the contract can explicitly authorize otherwise.
Can I use it in a directory without Git?
Yes. A persistent local folder can use artifact mode, completing items through artifact receipts under .nightshift/receipts/.
Is Nightshift Coding Shifts free?
The software is free, but Nightshift Coding Shifts calls an AI model, and you pay that model provider for usage (or run a local model).
Is Nightshift Coding Shifts open source?
Yes. The source code is on GitHub (orwa-mahmoud/nightshift) under the MIT licence.
Is Nightshift Coding Shifts still maintained?
Yes. The latest commit on GitHub is from 2026-11-19, within the last 30 days. The latest release is v0.37.2 (2026-11-19).
Who makes Nightshift Coding Shifts?
Nightshift Coding Shifts is published on GitHub by the account orwa-mahmoud. FollowAgents has not verified which person or company is behind that account.
What is Nightshift Coding Shifts's official website?
The project's GitHub repository lists https://nightshift.orwamahmoud.com as its website. That link is set by the repository owner; the source code itself is on GitHub (orwa-mahmoud/nightshift).
Was this page helpful?
View on GitHub ↗ Install ↓

Host ecosystems

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts