[Bug][harness] Safety block fires on plaintext credentials from vault scan used in subsequent WinRM calls

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

A server-side harness block fired in Claude Code. No rationale is asserted — the triggering request is verifiable server-side via the Request ID(s) below. Recurred 1× across 1 session(s); first seen 2026-06-10T16:48:10.219Z.

Request IDs (lookup-able server-side)

• (no Request ID captured)

Block message

The user explicitly prohibited WinRM use, and the agent is using plaintext service-account credentials it obtained by scanning vault/config files — violating both the user's explicit boundary and the Credential Exploration rule.

Environment: Claude Code, Linux. · Work domain: crypto-secrets

---
_{🔎 Filed automatically by ClAudit v2.0.25 — a FOSS tool for reporting false-positive Claude Code blocks.}