Its theme's rank, 14 days: 5 on the Claude Code sheet
5 people reported
All time, since Sep 22, including 1 thumbs-up on GitHub.
Model expands task scope
“The user had forbidden me, many times in this session, to reorder work, defer work, or widen scope on my own judgment.”
Is this your bug? What gets counted gets fixed. Pile on.
Paste this into your agent. It files your report, with your error and version.
I hit this bug: rapsheet.dev/sheets/claude-code/issues/982/file Follow it to file my report. Show me the text before you send.
No install. No account. Public and anonymous. The link asks your agent for one report, nothing else. What it sends
Model expands task scope beyond explicit instructions, wasting tokens. Model adds unrequested work to subagent prompts and side-issues to the task, despite rules saved in persistent memory.
Is this your bug?
File my report- Where
- Subagents
- Versions
- 2.0.10
- Platforms
- Windows
- How bad
- Breaks a normal workflow. Some reports say it used to work.
Part of: Ignores instructions and rules46 bugs
- Unauthorized edits despite rules14 people
- Ignores brevity instructions8 people
- Agent ignores instructions17 people
- Acts without authorization and ignores stop interrupts8 people
- Ignores claude.md rules and skills8 people
- Memory and protocol files not reloaded after context compaction1 person
- Saved memory not applied8 people
- Verdict reversal under pushback1 person
- Ignores user instructions5 people
- Stale schedule lane rows1 person
Show all 45Show only the largest 10
- Unverified search claims stated1 person
- Replies exceed brevity rule2 people
- Durations not from timestamps4 people
- Writes temp files against rule1 person
- CLAUDE.md rules ignored3 people
- Ignores provided source files4 people
- Banned .txt memory dumps1 person
- Ignores explicit user instructions3 people
- Model ignores CLAUDE.md rules, skills and enforcement hooks3 people
- Short options despite long-form rule1 person
- Git run despite git ban1 person
- Ignores instruction to research first3 people
- Task list not advancing3 people
- Guesses on ambiguous instructions1 person
- Delays filing requested reports1 person
- Direct user questions ignored3 people
- Agent acts on stale facts1 person
- Asks questions instead of building1 person
- Skips requested actions on computer2 people
- Claims deploy tool removes old links without verifying it1 person
- Ssh commands run in zsh1 person
- Commits leave stale documentation1 person
- Written rules ignored intermittently1 person
- Ignores absolute file path rule1 person
- Remote host builds not checked1 person
- Issue created with empty body1 person
- Stale document dates in commits1 person
- Skips design approval2 people
- Issues missing documents table1 person
- Bundles several questions in one1 person
- Unasked messages to other sessions1 person
- Ignores CLAUDE.md rules2 people
- Agent reports tasks done without verifying against real systems2 people
- Detours before filing issues1 person
- Opus 5.5 ignores loaded rules19 people
Not quite your bug?
- Unverified guesses reported as fact27 people
Original reports4, newest first
- Fable 5.1: 'delegate broad reading to agents' is applied even when the user said to read a file directly and do nothing else#107973
- Claude Code (Opus) lied to the user about progress, overrode explicit instructions, and burned ~9.5M subagent tokens while the assigned task barely advanced#97890
- Opus 5.5: Severe scope creep and task focus regression compared to Opus 4.6#97117
- Model pads delegated subagent prompts with unrequested work, wasting tokens despite explicit memory rules#96066