# Nightshift

> Autonomous overnight codebase improvement agent for Claude Code. Run it before bed, wake up to production-ready fixes.

- **URL**: https://www.freshcrate.ai/projects/Nightshift
- **Author**: Recusive
- **Category**: Testing
- **Latest version**: `v0.0.7` (2026-04-05)
- **License**: MIT
- **Source**: https://github.com/Recusive/Nightshift
- **Homepage**: https://orbit.build/nightshift
- **Language**: Python
- **GitHub**: 1 stars, 1 forks
- **Registry**: github
- **Tags**: `ai-agent`, `anthropic`, `automation`, `claude`, `claude-code`, `code-quality`, `developer-tools`, `overnight-agent`, `python`

## Description

Autonomous overnight codebase improvement agent for Claude Code. Run it before bed, wake up to production-ready fixes.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v0.0.7` | 2026-04-05 | High | ## Highlights  - **Prompt injection protection**: Target repo instruction files are pre-read and wrapped in untrusted context blocks - **Prompt self-modification guard**: Checksums 7 prompt/control files before/after each cycle, flags tampering - **Cost tracking**: Per-session token parsing with model-specific pricing, budget ceiling, cumulative ledger - **Self-evaluation loop**: 10-dimension scoring of nightshift runs against real repos - **Loop 2 complete**: Feature summary, production-readine |
| `v0.0.6` | 2026-04-05 | Medium | ## Highlights  - **Repo profiling**: Detects language, framework, package manager, test runner, and structure - **Feature planner**: Builds structured plans from natural language descriptions with topological wave sorting - **Task decomposer**: Converts plans into sub-agent work orders with complete prompts - **Sub-agent spawner**: Spawns and manages codex/claude agents per work order with retries - **Wave integrator**: Merges sub-agent work, runs tests, diagnoses failures, spawns fix agents - * |
| `v0.0.5` | 2026-04-04 | Medium | ## Highlights  - **Multi-repo support**: New `nightshift multi` subcommand runs hardening shifts across multiple target repos in a single command. Validates all repos upfront, runs sequentially, prints aggregate summary. - **Loop 1 complete**: With multi-repo, all 22 Loop 1 components are done (100%). The hardening loop is feature-complete.  ---  # v0.0.5 -- Multi-Repo  **Released**: 2026-04-03 **Tag**: `v0.0.5` **Status**: Released  Multi-repo support: run hardening loops across multiple target |
| `v0.0.4` | 2026-04-04 | Medium | ## Highlights  - **Category balancing**: Agents now get steered toward unexplored fix categories after a configurable cycle threshold, preventing lopsided shifts that only fix one type of issue - **Shift-log tolerance**: Verification no longer rejects cycles where agents commit shift-log updates separately from fixes -- matches real agent behavior - **Runner hardening**: Worktree validation, repo checkout validation, forbidden command detection, high-signal focus paths, and npm lockfile preventi |
| `v0.0.3` | 2026-04-04 | Medium | ## Highlights  - **Diff scorer**: Every cycle's changes scored 1-10 based on production impact. Low-scoring cycles are reverted automatically. - **State injection**: Agents now receive a summary of what prior cycles found, preventing redundant exploration. - **Test incentives**: Agents that don't write tests get escalating pressure after 3 cycles. - **Backend forcing**: Full-stack repos get steering directives when agents fixate on frontend code. - **Security fix**: `merge_config()` now deep-mer |
| `v0.0.2` | 2026-04-03 | Medium | Replaces bash orchestration with a Python control plane. Adds runner-enforced guard rails, structured state, and pluggable agent adapters. Nightshift can now use Codex or Claude — same pipeline, you just pick.  ## Highlights  - **nightshift/ package** — 10-module Python orchestrator replacing the monolithic script - **Equal agent adapters** — Codex and Claude are first-class. `resolve_agent()` priority: CLI flag > config > interactive picker - **Runner-enforced guard rails** — Max fixes, blocked |
| `v0.0.1` | 2026-04-03 | Medium | ## Nightshift v0.0.1 — Initial Beta  The first release of Nightshift, an autonomous overnight codebase improvement agent for Claude Code.  ### What's included  - **SKILL.md** — The core skill with discovery strategies, gotchas section, safety rails, shift log template, and worktree isolation - **run.sh** — Multi-cycle overnight runner (8-10 hours, fresh Claude sessions per cycle) - **test.sh** — 4-cycle test runner for quick validation  ### How it works  - Runs in an isolated git worktree — your |

## Dependency audit

- **Score**: 100/100
- **Total deps**: 0
- **Resolved**: 0
- **Unresolved**: 0
- **License conflicts**: 0
- **Warnings**: 0
- **Scanned**: 2026-05-04

## Citation

- HTML: https://www.freshcrate.ai/projects/Nightshift
- Markdown: https://www.freshcrate.ai/projects/Nightshift.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/Nightshift/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._