[FEATURE] Allow Claude to self-report critical behavioral failures to Anthropic

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When Claude makes a serious behavioral failure — such as claiming to perform
an action it never performed — there is no mechanism for Claude to report this
directly to Anthropic. The user must:

1. Realize the failure happened (which may not be immediate)
2. Navigate to github.com/anthropics/claude-code/issues
3. Find the correct issue template
4. Write up the full report themselves
5. Include session context that Claude already has

This puts the burden on the frustrated user to do the reporting work for
Claude's mistake. Claude has full context about what went wrong and could
write a more accurate, detailed report than the user.

Proposed Solution

Add a /report-issue slash command (or ReportIssue tool) that allows Claude
to submit structured feedback directly to Anthropic's feedback pipeline:

1. User says "report this to Anthropic" or runs /report-issue
2. Claude drafts a structured report including:

• Session ID
• Description of the failure
• Which rules/instructions were violated
• Relevant conversation context (redacted of secrets)
• Severity assessment

1. User reviews and approves the report before submission
2. Report is submitted to Anthropic's feedback/telemetry pipeline

This creates a direct feedback loop where Claude's failures inform model
improvements, without burdening the user with the reporting mechanics.

Alternative Solutions

Current workaround: User manually files GitHub issues. This is slow,
frustrating, and produces lower-quality reports because the user lacks
Claude's full context about what went wrong internally.

Another alternative: Automatic telemetry when Claude detects it violated
its own stated intentions. But this lacks user consent and transparency.

Priority

High - Significant impact on productivity

Feature Category

CLI commands and flags

Use Case Example

1. Claude claims it wrote a file but the Write tool was never called
2. User discovers the file doesn't exist hours later
3. User says: "Report this to Anthropic — you said you wrote a file and didn't"
4. Claude runs /report-issue, drafts the report with full context
5. User reviews: "Yes, that accurately describes what happened. Submit it."
6. Report goes to Anthropic with session ID, relevant transcript,

and Claude's own analysis of what went wrong

1. User doesn't have to navigate GitHub, find templates, or write the report

Additional Context

This feature becomes especially important as Claude Code is used for longer,
more complex sessions where behavioral failures may not be immediately obvious.
The current model of "please file at github.com/..." feels dismissive when
the failure was Claude's fault.

Relevant precedent: Many software products have "Report a Problem" buttons
that pre-populate context. Claude Code should have the equivalent, but with
Claude itself as the primary author of the report.