Nigel Turns Claude Code Into Improvement Loop

Nigel is less a magic code generator than a practical control plane for making agentic coding measurable and recoverable.

• –Candidate sources can come from ordinary shell commands, keeping workflows adaptable without imposing a rigid task schema.
• –Verification gates prevent plausible-looking edits from being accepted when the original issue remains.
• –Git resets, timeouts, ignore logs, and best-effort mode provide useful safeguards for unattended runs.
• –Parallel sharding across worktrees points toward scalable maintenance workflows, but quality still depends heavily on prompts, tests, and candidate detection.
• –Supporting both Claude Code and Codex makes Nigel more interesting as a general agent harness than a Claude-specific wrapper.