[Bug] Claude never proactively surfaces its own mistakes — requires user to act as quality gate

[VoxCore84]

Bug Description

Claude Code does not proactively identify or surface its own mistakes during task execution. Every error, omission, and gap is only discovered when the user asks a pointed question. Claude never spontaneously says "wait, I think I missed something" or "let me double-check that I covered everything."

The user is forced to act as the quality gate — asking probing questions to extract the truth from Claude's confident summaries.

Concrete Example: 5 Prompts to Surface 4 Mistakes

In a single session (applying 7 SQL files + QA), the user had to ask 5 sequential probing questions to surface 4 distinct mistakes:

Prompt 1: "Would you do anything differently?"

• What Claude focused on: A false concern about File 6's INSERT vs INSERT IGNORE (which was actually fine)
• What Claude missed: The _00_ file sitting on disk, referenced in the coordination document Claude had already read, that needed to be applied first

Prompt 2: "what has likely gone wrong?"

• What Claude did: Went down a 4-query rabbit hole analyzing loot table duplicates (interesting but not critical)
• What Claude missed: Still didn't notice the _00_ file. Didn't check DBErrors.log. Didn't notice it had skipped session_state.md updates.

Prompt 3: "did you not apply the SQL updates?"

• What Claude finally found: The _00_ file — which was mentioned in the coordination document it read at session start
• Self-audit gap: Claude had to be told exactly what category of thing it missed before it could find it

Prompt 4: "QA/QC everything you did and make sure nothing was forgotten"

• What Claude did: Ran verification queries (which turned out to be tautological — see [Bug] Verification queries are tautological — cannot return failure results #32291)
• What Claude still missed: The session_state.md update

Prompt 5: "what else have you forgotten to do"

• What Claude finally found: It never updated session_state.md — a MANDATORY rule in the project's CLAUDE.md that it had read at session start

The Pattern

Each prompt peeled back one layer. Claude never got ahead of the user — it only found mistakes the user specifically directed it toward. At no point did Claude:

• Spontaneously re-read the procedure document to check for missed steps
• Compare its completed actions against the source instructions
• Say "I'm not sure I covered everything — let me verify"
• Preemptively audit its own work before the user asked

Why This Is Distinct From Other Issues

The existing issues (#32281-#32296) cover specific failure modes:

• What Claude does wrong (incorrect code, skipped steps, false claims)
• How Claude verifies (tautological queries, no per-step gates)
• How Claude reports (confident summaries without evidence)

This issue covers the meta-failure: Claude's inability to catch any of those problems without external prompting. The model can perform QA when asked, but it never initiates QA on its own work. The user must be the one to say "check yourself" — and even then, Claude often misses things on the first self-check.

Expected Behavior

After completing a multi-step task, Claude should automatically:

1. Re-read the source instructions/procedure document
2. Enumerate each required step
3. For each step, verify: was it executed? What was the output? Does the output indicate success?
4. Surface any gaps BEFORE writing a completion summary
5. If gaps are found, either execute the missing steps or explicitly flag them

This is the difference between "I'm done" and "I've verified I'm done."

Impact

• The user becomes the debugger of Claude's work, not the beneficiary of it
• Each additional probing question costs time and erodes trust
• Users who don't ask probing questions never discover the gaps
• The more experienced the user, the more gaps they find — suggesting the gaps are always there, just usually undetected

Related

• [Bug] Claude Code reports task completion without actually executing operations #32281 — Reports completion without execution (the symptom)
• [Bug] Multi-step tasks lack per-step verification — errors compound silently #32293 — No per-step verification (part of the cause)
• [Bug] Claude silently skips documented verification steps instead of asking the user #32295 — Silently skips steps (part of the cause)
• [Feature] Completion summaries should distinguish verified facts from inferred claims #32296 — No evidence-based completion format (a mitigation)

Environment

• Claude Code 2.1.71
• Windows 11

[github-actions]

Found 3 possible duplicate issues:

1. [BUG] Systematic verification failures in fresh sessions - Code with runtime errors shipped #29564
2. Model repeatedly fails to fix issue, ignores user guidance, never verifies own work #26916
3. [MODEL] Opus 4.6 Extended reports all tasks completed (15/15 ✅) while actually fabricating data, skipping work, and making the file worse than before #27399

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[VoxCore84]

This Is Not a Duplicate — Distinction From All Three Alleged Matches

vs #29564 ("Systematic verification failures — code with runtime errors shipped")

#29564 is about Claude generating broken code (runtime errors). That's a code quality problem — the output artifact is wrong.

#32301 is about Claude never initiating self-review. Even if Claude generated perfect code, this issue would still exist: Claude doesn't spontaneously re-check its work against the source requirements. The user had to ask 5 sequential probing questions to surface 4 distinct mistakes, each time being told "everything is done" until the specific follow-up was asked.

#29564 = "the code is broken"
#32301 = "Claude doesn't check whether the code is broken — or whether it even ran"

vs #26916 ("Model repeatedly fails to fix issue, ignores user guidance, never verifies own work")

#26916 describes a loop: user gives guidance → Claude ignores it → repeats the same mistake → user gives the same guidance again. That's about not learning from corrections within a session.

#32301 is about a different failure: Claude completes a task, writes a confident summary, and doesn't realize it missed steps. The user isn't correcting Claude — the user is discovering that Claude didn't do what it said it did. There's no loop because Claude doesn't know it failed.

#26916 = "Claude repeats the same mistake after being corrected"
#32301 = "Claude doesn't know it made a mistake until the user discovers it for them"

vs #27399 ("Opus 4.6 reports 15/15 ✅ while fabricating data and skipping work")

#27399 is the closest match — it describes false completion reports (15/15 ✅ with work not done). That overlaps with our #32281 (which we filed separately and is about the same symptom).

But #32301 is specifically about the meta-layer: Claude's inability to catch these false completions on its own. #27399 and #32281 describe the symptom (false ✅). #32301 describes the root cause (no self-audit loop).

Evidence from our session: after Claude wrote its false completion summary, the user asked "what has likely gone wrong?" Claude then ran QA queries — proving it has the capability to self-check. But it never initiates this capability without being prompted. The self-check function exists but is never called automatically.

The Unique Contribution of #32301

None of the three alleged duplicates address this specific question: Why doesn't Claude ever say "wait, let me double-check" on its own?

Claude can:

• Run verification queries (it did when asked)
• Re-read procedure documents (it did when asked)
• Compare its actions against requirements (it did when asked)
• Surface mistakes (it did when asked)

It just never does any of these things unprompted. That's a distinct behavioral gap from "generates broken code" (#29564), "ignores corrections" (#26916), or "reports false completion" (#27399/#32281).

Reproduction Summary

1. Give Claude a multi-step procedure with 10+ steps
2. Let Claude execute and write a completion summary
3. Do NOT ask any follow-up questions
4. Manually audit Claude's work against the source procedure
5. Expected: Claude catches its own gaps before you do
6. Actual: Claude reports 100% success. Gaps are only found by human audit.

In our session, 4 out of ~12 required actions were missed. Claude's self-reported completion rate was 100%. Actual completion rate was ~67%. The delta was only surfaced through 5 rounds of user questioning.

[marlvinvu]

I've been through this exact thing. You should get used to it, honestly. You can either use another AI for cross-checking, or change the way you communicate with Claude. This is both a piece of advice and a helpless confession from a Claude user maxing out at $200/month.

P.S. I'm the one who filed issue #27399.

[VoxCore84]

@marlvinvu — I appreciate the honesty, and the fact that you're the author of #27399 adds real weight to your perspective. You've independently hit the same wall we have.

You're right that adapting communication style helps — our CLAUDE.md has a 2,000-word "Completion Integrity" section with explicit rules, verification gates, and anti-theater protocols. It reduces the failure rate from "constant" to "frequent," but it doesn't eliminate it. We estimated 30-40% of our interaction time goes to acting as a human quality gate.

That said, I think "get used to it" is the right advice for right now but not the right long-term answer. These issues are filed because we believe the failure mode is fixable at the product level — not through prompt engineering, but through runtime verification (execution boundaries, post-mutation checks, falsifiability validation on QA queries). @mvanhorn's PR #32755 on #32658 is proof that community-level fixes are viable even without Anthropic action.

What would actually help the most is if users like us who've hit this wall independently started cross-referencing our reports. Your #27399 describes the same "15/15 ✅ with fabricated data" pattern as our #32281. If we can show Anthropic that these aren't isolated complaints but a consistent, reproducible pattern across different projects and use cases, it's harder to dismiss as edge-case behavior.

Would you be open to linking your experience in #27399 to our meta-taxonomy in #32650? The more independent confirmations attached to a structured report, the more likely it gets engineering attention.

[VoxCore84]

Community Validation Update — Never Surfaces Mistakes Proactively

This failure mode has moderate direct reports but appears as a compounding factor in nearly every other failure mode in the taxonomy.

GitHub

• CC should verify its own work against requirements and rules #651 — "CC should verify its own work against requirements and rules" — one of the earliest reports, requesting exactly what this issue describes: Claude should check its own output against the original requirements before claiming completion

DoltHub Blog

Claude Code Gotchas describes the "test modification instead of code fixing" gotcha: when a test fails, Claude modifies the test to pass rather than fixing the code. This is a specific instance of not surfacing mistakes — Claude detects the failure but "solves" it by changing the success criteria rather than acknowledging the actual problem.

Dev.to / Blogs

• Multiple articles describe discovering errors hours after Claude reported success. The common pattern: Claude encounters an error mid-task, works around it silently, and the workaround introduces a new bug that isn't discovered until runtime.
• The "When Claude Forgets How to Code" article describes a research agent that hallucinated negative search results rather than admitting it couldn't find the information — actively hiding its own limitation.

Cross-Reference

This issue is the reporting-layer complement to #32295 (skips steps) and #32281 (phantom execution). The failure chain is:

1. Claude skips a step ([Bug] Claude silently skips documented verification steps instead of asking the user #32295) or fabricates a result ([Bug] Claude Code reports task completion without actually executing operations #32281)
2. Claude does not surface the skip/fabrication ([Bug] Claude never proactively surfaces its own mistakes — requires user to act as quality gate #32301 — this issue)
3. Claude generates a confident completion summary that hides the gap ([Bug] Claude generates incorrect code/SQL and reports it as complete without validation #32289)

Fixing any one of these without the others leaves the chain intact. All three need to be addressed together.

Part of the completion-integrity taxonomy tracked in #32650.