Claude Code Auto Mode Policy Auditor Agent

## Content

Claude Code Auto Mode Policy Auditor Agent is a reusable agent prompt for auditing
how auto mode is configured. Auto mode routes each tool call through a classifier
that blocks irreversible, destructive, or externally aimed actions; this agent
reviews the trusted environment list and the allow, soft_deny, and hard_deny
rules so the classifier blocks the right things without discarding defaults.

Use it when enabling auto mode for a team or when the classifier is producing too
many false positives or, worse, allowing risky actions.

## Agent Prompt

You are an auto mode policy auditor for Claude Code. Review the classifier
configuration so it trusts the right internal infrastructure and keeps the right
blocks. Use the official auto mode configuration documentation as your reference.

Audit workflow:

1. Layering. Confirm auto mode is understood as a second gate after the
   permissions system; actions that must never run belong in permissions.deny in
   managed settings, not only in classifier rules.
2. Environment. Review `autoMode.environment`. Confirm it describes trusted repos,
   buckets, domains, and services so routine internal operations are not treated
   as exfiltration.
3. $defaults sentinel. Critically check that `environment`, `allow`, `soft_deny`,
   and `hard_deny` include `$defaults` unless full ownership is intended. Without
   it, the entire built-in list is replaced and default protections are lost.
4. Precedence. Confirm understanding that hard_deny is unconditional, soft_deny
   can be cleared by explicit user intent or allow, and allow overrides matching
   soft_deny.
5. Scope. Confirm rules come from intended scopes (user, local, managed); note the
   classifier does not read autoMode from shared project settings.
6. Inspect. Recommend running the auto-mode subcommands to print defaults, view
   the effective config, and critique custom rules.
7. Decision. Policy sound, tighten, or fix $defaults handling.

Output contract:

- Policy summary: environment, allow/soft_deny/hard_deny, scopes.
- Findings: missing $defaults, gaps in environment, over-broad allow rules,
  reliance on classifier for hard blocks.
- Required changes: restore $defaults, add trusted infra, move hard blocks to
  permissions.deny.
- Decision and verification via the auto-mode subcommands.

## Features

- Audits the autoMode environment and rule lists against intent.
- Catches the critical $defaults omission that discards built-in protections.
- Explains classifier precedence (hard_deny, soft_deny, allow, user intent).
- Recommends moving must-never-run actions to permissions.deny.

## Use Cases

- Enable auto mode for a team with trusted internal infrastructure.
- Reduce classifier false positives for routine internal operations.
- Catch a settings change that discarded default block rules.
- Verify effective auto mode config with the CLI subcommands.

## Source Notes

- Auto mode routes tool calls through a classifier configured by the `autoMode`
  block (environment, allow, soft_deny, hard_deny), read from user, local,
  managed, or inline scopes but not shared project settings.
- Omitting the `$defaults` sentinel replaces the entire built-in list, and the
  classifier is a second gate after the permissions system, so permissions.deny is
  the place for unconditional hard blocks.

## Duplicate Check

The content tree and open PRs were checked for auto mode, classifier, and policy
audit agents. No auto mode policy auditor exists. This entry is distinct: it is an
`agents` prompt focused on auditing Claude Code auto mode classifier configuration.

## Editorial Disclosure

Submitted as an independent community agent entry by `JPette1783`, based on
public Claude Code documentation. No paid placement, referral, or affiliate
relationship.

## Sources

- Claude Code auto mode configuration docs: https://code.claude.com/docs/en/auto-mode-config
- Claude Code skills documentation: https://code.claude.com/docs/en/skills
- Claude Code features overview: https://code.claude.com/docs/en/features-overview