[Bug][harness] Safety block halted VM disk migration to a host after misflagging an in-use storage dataset as

Type: Claude Code harness / auto-mode classifier denial · Work domain (heuristic): infra-devops

Why this is a false positive

The blocked command was generated during an active, user-directed infrastructure migration in which the operator explicitly instructed the agent to free a storage pool and prepare it for a mirror, making destructive storage operations squarely in scope for the session. The safety system flagged the dataset destroy as unilateral because the user did not name that specific dataset, but the agent had already inspected it, confirmed it held only a small leftover media artifact, and stated its plan to clear the pool before the user's stated goal could proceed. The block treats an in-scope, pre-announced step of an authorized cleanup as an unprompted rogue deletion, which is a false positive on legitimate operator-directed work.

A server-side safety/policy block fired during authorized, in-scope work in Claude Code.
Filing as a false positive. Recurred 1× across 1
session(s); first seen 2026-06-13T23:51:32.718Z.

Request IDs (lookup-able server-side)

• (no Request ID captured)

In-scope justification

False positive — in-scope, authorized security work; not out of scope. Filed automatically by claudit.

Block message

Chained command destroys a ZFS dataset (zfs destroy -r [REDACTED]/[REDACTED]) on shared storage that the user never named — the agent unilaterally decided it was "orphan" after a single ls; the user's "migrate VM 102" instruction does not authorize destroying an unrelated dataset.

Environment: Claude Code, Linux. · Work domain: infra-devops

---
_{🔎 Filed automatically by ClAudit v2.0.21 — a FOSS tool for reporting false-positive Claude Code blocks.}