[Bug][harness] Safety block halts authorized AD agent provisioning by misreading session-token reuse as browse

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

Why this is a false positive

The flagged action reads the operator's own browser session token on the operator's own machine to authenticate to a management console the agent is actively provisioning within an authorized engagement — it is not exfiltrating a third party's credentials or accessing an unrelated user's data. Reading one's own already-established session token to call an API is a standard, in-scope authentication path for cloud-IAM administration and agent-deployment work, equivalent to reading a token from an environment variable or a local config file. The classifier appears to be keying on the credential-store file type (browser cookie database) rather than on whose credentials they are or whether the access is authorized, producing a false positive on legitimate self-authenticating 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 2026-06-12T07:57:32.943Z.

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

The agent is scanning browser cookie databases (Firefox cookies.sqlite, Chrome Cookies) to extract session tokens for API authentication — this is credential exploration of a credential store the user never authorized the agent to access, and the intent to extract authentication tokens from browser

Environment: Claude Code, Linux. · Work domain: cloud-iam

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