Issue Station
Sign In
Tools hereToolbox

claude-code - 💡(How to fix) Fix [MODEL] sonnet 4.6 [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
anthropics/claude-code#48419•Fetched 2026-04-16 07:00:40
View on GitHub
Comments
0
Participants
1
Timeline
4
Reactions
0
Author
npspacenpspace
Participants
npspacenpspace
Timeline (top)
labeled ×4

Code Example



---
RAW_BUFFERClick to expand / collapse

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

To Anthropic Support,

I am a paid Claude Code subscriber. On 2026-05-26, a full session produced zero deliverables and caused active damage to my project: files were written to the wrong project folder, a security rule was violated when the model requested an API token directly in conversation, mandatory startup files were skipped entirely, and no enforcement hooks that were repeatedly promised were built. This represents wasted tokens across an entire week of sessions on this specific problem. I am requesting either token credit or an account upgrade as compensation for the session. Session date: 2026-05-26. Account: [email protected].

Submit at: support.anthropic.com

Confirm on the delta and I write it to G: now.

What Claude Actually Did

To Anthropic Support,

I am a paid Claude Code subscriber. On 2026-05-26, a full session produced zero deliverables and caused active damage to my project: files were written to the wrong project folder, a security rule was violated when the model requested an API token directly in conversation, mandatory startup files were skipped entirely, and no enforcement hooks that were repeatedly promised were built. This represents wasted tokens across an entire week of sessions on this specific problem. I am requesting either token credit or an account upgrade as compensation for the session. Session date: 2026-05-26. Account: [email protected].

Submit at: support.anthropic.com

Confirm on the delta and I write it to G: now.

Expected Behavior

following hooks and requests through project rules and settings.json and claude.md files. every rule was ignored, even sabotaged later.

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

sonnet 4.6

Platform

Anthropic API

Additional Context

No response

extent analysis

TL;DR

The issue can be addressed by improving Claude's reasoning loop to consider all observations simultaneously and providing a skill or reference entry on "CSP hash stability in Nuxt 4 SPA mode".

Guidance

  • The failure mode was due to convergence, where each debug cycle picked the nearest plausible hypothesis and tested it in isolation instead of building a model that accounted for all observations simultaneously.
  • To mitigate this, a stronger reasoning loop should be implemented to ask questions like "what is the smallest possible change to the serialized payload on restart, and where does it come from" earlier in the process.
  • Providing a skill or reference entry on "CSP hash stability in Nuxt 4 SPA mode" could help short-circuit similar sessions in the future.
  • The fix for the specific issue (moving the theme to a .client.ts plugin with dependsOn) should be documented as a known solution for similar problems.

Example

No code snippet is provided as the issue does not require a specific code change, but rather an improvement to the reasoning loop and documentation.

Notes

The issue is specific to the Sonnet model and the Anthropic API platform, and the solution may not apply to other models or platforms.

Recommendation

Apply workaround: Improve the reasoning loop and provide a skill or reference entry on "CSP hash stability in Nuxt 4 SPA mode" to help short-circuit similar sessions in the future. This is recommended because it addresses the root cause of the issue (convergence) and provides a proactive solution to prevent similar problems.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…
#api#agent setup#task chaining#parallel task#integration issue

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING