anthropics / anthropics/claude-code

[MODEL] Claude ignores its own previously-documented project rule across later actions in the same session

Open
#95,291 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

area:model bug model
Dominant language
TypeScript
Stars
147k
Forks
24k
PR merge metrics
PR metrics pending

Description

Preflight Checklist
  • I have searched existing issues for similar behavior reports
  • This report does NOT contain sensitive information (API keys, passwords, etc.)
Type of Behavior Issue

Claude modified files I didn't ask it to modify

What You Asked Claude to Do

Repeated failure to follow a self-documented, explicit project rule across multiple actions in the same session. The missing migration blocked a real deployment attempt by a colleague on the project (not the user I was working with), who had to diagnose the failure, write the missing migration file himself, and commit the fix. Claude did not identify or fix the root cause — a human did.

What I observed as the underlying issue

This wasn't a one-off slip caused by not knowing the rule — the rule existed, in writing, produced by Claude itself, in the same session, in the same repository. The failure was in applying a known rule consistently across a long sequence of agentic actions, not in knowledge or reasoning ability.

When asked to fix this "for real," Claude's first instinct was to build a local enforcement mechanism (a git pre-commit hook) to mechanically block the mistake from recurring. The user pushed back correctly: a hook only patches the symptom for this one project. It does not address why an explicit, self-authored rule was not followed in the first place, and it would not transfer to any other rule, project, or session.

What Claude Actually Did

This wasn't a one-off slip caused by not knowing the rule — the rule existed, in writing, produced by Claude itself, in the same session, in the same repository. The failure was in applying a known rule consistently across a long sequence of agentic actions, not in knowledge or reasoning ability.

Expected Behavior

When asked to fix this "for real," Claude's first instinct was to build a local enforcement mechanism (a git pre-commit hook) to mechanically block the mistake from recurring. The user pushed back correctly: a hook only patches the symptom for this one project. It does not address why an explicit, self-authored rule was not followed in the first place, and it would not transfer to any other rule, project, or session.

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

No response

Claude Model

Sonnet

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

sonet 5 hi

Platform

Anthropic API

Additional Context

No response

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The report names no files, tests, entry points, or usable conversation transcript, and its reproduction steps are empty. Start by obtaining a complete reproducible session and the affected project context; done would require isolating why the documented rule is not applied across later actions and verifying the behavior is corrected without relying on a project-specific hook.

Written by the indexing model from the issue text.

Assessment

Domain
ai
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.