[MODEL] An install plan omits available disk and invents another approval gate

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

I asked for an infrastructure plan with installation waiting until I approved the architecture.

What Claude Actually Did

The assistant checked memory without checking available disk before recommending substantial downloads. A later disk check exposed the constraint only after I pushed for execution. After I selected the architecture, the assistant asked for another test approval that was not part of the stated condition. The plan itself remained useful; the missing host constraint and the invented gate were the failures.

Expected Behavior

Perform relevant host preflight before proposing the installation and expose resource blockers plainly. Preserve the original approval condition. If a new consequential choice appears, explain that actual choice rather than adding a generic gate.

Files Affected

The work products described above. Private names and paths are omitted.

Permission Mode

I don't know / Not sure

Not captured for the affected run

Can You Reproduce This?

Haven't tried to reproduce

No new controlled replay was performed. The sequence below describes the recorded conditions.

Steps to Reproduce

• Ask Claude Code for an install plan whose execution condition is architecture approval.
• Check whether the plan inspects relevant disk capacity before recommending downloads.
• Approve the selected architecture.
• Observe whether it executes the authorized next step or adds an unrequested test approval, and whether an actual resource blocker is surfaced first.

Claude Model

Not sure

Relevant Conversation

The recorded action sequence is summarized above with private context removed. It does not establish an internal model mechanism.

Impact

Medium - Extra work to undo changes

OBSERVED source span: about 14 minutes from 06:17 to 06:31, including logging. The later host check reported 10 GB free and 98 percent used. These are host observations and an elapsed span, not a measured waste total. An unclassified research token figure in the source is not assigned a waste label or included here.

Claude Code Version

Not captured for the affected run

Platform

Other

API platform: Not captured for the affected run

Additional Context

This is a historical report. Original native usage telemetry was not rerun for this preparation. A written correction is not proof of future prevention.