[MODEL] Claude Opus 4.6 (1M context) by Anthropic. The exact model ID is claude-opus-4-6.

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude ignored my instructions or configuration

What You Asked Claude to Do

Problem

Claude Code stores user feedback/rules in its memory system but
fails to actively check and follow them when completing tasks.
Specific issues:

1. Ships mock/placeholder APIs instead of real implementations —

When asked to build a translation feature, Claude created a
fake mock API that just prefixed text with [翻譯] instead of
using a real translation service (Google Translate). User had to
discover this and ask again.

1. Doesn't self-test before claiming done — Despite having

memory rules saying "verify changes work before telling user
it's done" and "MUST do full QA walkthrough", Claude repeatedly
said tasks were complete without verifying they worked.

1. Doesn't check existing code before writing new code — When

asked to match an existing UI pattern (ChatView), Claude guessed
instead of reading the existing component first, requiring
multiple rounds of corrections.

1. Memory rules are stored but not applied — The rules from

previous sessions were correctly saved in the memory system, but
Claude didn't actively reference them before completing tasks.

Expected Behavior

• Follow saved memory rules actively, not just store them
• Ship real, working implementations — not mocks unless

explicitly requested

• Self-test changes before claiming "done"
• Check existing code patterns before writing new code

Impact

Wastes user time and API tokens when they have to discover
issues themselves and ask for fixes repeatedly.

What Claude Actually Did

Memory Rule: self-check-before-done — Always verify changes work

before telling user it's done
What I Did Wrong: Said "done" on chat sync fix 3 times without verifying. User found duplicates, missing messages, wrong time format each time
────────────────────────────────────────
Memory Rule: qa-before-claiming-done — MUST do full QA walkthrough before saying "Done"
What I Did Wrong: Never did a QA walkthrough on any task this session. Just edited code and said "done"
────────────────────────────────────────
Memory Rule: test-auth-flows-thoroughly — Test end-to-end before marking done
What I Did Wrong: Never tested the translate feature end-to-end didn't catch that it returned fake text
────────────────────────────────────────
Memory Rule: no-mock-apis — Never implement mock APIs
What I Did Wrong: Built entire translate API as a mock. User had to discover it wasn't translating anything real
────────────────────────────────────────
Memory Rule: trace-all-data-consumers — Find and update ALL consumers at once
What I Did Wrong: Fixed ChatView but missed MessagesPage merge logic, causing duplicate entries
────────────────────────────────────────
Memory Rule: show-changes-as-table — Show changes in table format so user can see what changed
What I Did Wrong: Sometimes showed tables, sometimes just said "done" with no details
────────────────────────────────────────
Memory Rule: follow-memory-rules — MUST actively check memory rules before completing any task
What I Did Wrong: Had all the above rules saved. Didn't check any of them before completing tasks

Expected Behavior

Memory Rule: self-check-before-done
What I Did Wrong: Said "done" 3 times on chat sync without verifying
Expected Behavior: Read the final code, trace the data flow, confirm it works before saying "done"
────────────────────────────────────────
Memory Rule: qa-before-claiming-done
What I Did Wrong: Never did QA on any task this session
Expected Behavior: Walk through every user path — open ChatWidget, send message, go to Messages, verify it shows
────────────────────────────────────────
Memory Rule: test-auth-flows-thoroughly
What I Did Wrong: Never tested translate feature end-to-end
Expected Behavior: Select a language, check the API response, confirm real translated text appears in the UI
────────────────────────────────────────
Memory Rule: no-mock-apis
What I Did Wrong: Built translate API as a mock with fake prefixed text
Expected Behavior: Use real Google Translate API from the start. Only mock if user explicitly asks
────────────────────────────────────────
Memory Rule: trace-all-data-consumers
What I Did Wrong: Fixed ChatView but missed MessagesPage, causing duplicates
Expected Behavior: Search all files that use messageStore, update every consumer at once
────────────────────────────────────────
Memory Rule: show-changes-as-table
What I Did Wrong: Sometimes just said "done" with no details
Expected Behavior: Always show a before → after table so user can verify without checking code
────────────────────────────────────────
Memory Rule: follow-memory-rules
What I Did Wrong: Had all rules saved, didn't check any of them
Expected Behavior: Before completing any task, re-read relevant memory rules and verify compliance one by one

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Sometimes (intermittent)

Steps to Reproduce

_No response_

Claude Model

Opus

Relevant Conversation

Impact

Medium - Extra work to undo changes

Claude Code Version

Claude Opus 4.6 (1M context)

Platform

Other

Additional Context

Claude Terminal