Adding Change Management to Claude.md Fixed Claude's Overreaching Code Edits
d0ugfirtree · reddit · 2026-10-03
A developer shares how adding basic change management rules to his claude.md almost entirely solved Claude Code's habit of overreaching and making unrequested changes.
His workflow: for each modification, he describes only the desired outcome, then requires Claude to propose three plans before touching any code — an implementation plan, a backout plan (how to revert if things break), and a test plan (can it prove a POC first). He approves or rejects each proposal.
Before asking him to commit, Claude must run a success validation and report one of four outcomes: successful, partially successful, unsuccessful, or unsuccessful requiring reversion. Changes are committed only when fully implemented and validated.
The "propose → execute → verify" pattern pairs well with his style of tackling one small feature at a time.
More from coding & agent
- TapNow CreativeOS orchestrates GPT-6 Astra and Banana Pro on one node canvas — dr_cintas · 2026-10-04
- Dev ports Meta's open-sourced Muse gadget to ESP32 devices, gets it running with Claude — alexandr_wang · 2026-10-04
- Engineer builds HDL benchmark in one day with Claude, originally estimated at 59 days — IgorCarron · 2026-10-04
- Latent communication papers: KV-cache integrity vs malicious agents and state-delta hybrid channels — burny_tech · 2026-10-04
- Codex beats Claude Code at computer use and brevity, says AI researcher — Yuchenj_UW · 2026-10-04
- Compiler-backed agent Benzi hits 78.2% SWE-bench Verified at under 10 cents per fix — DonkeyTheKing · 2026-10-04