Skip to content

Rule Reference

SEC362lintai-ai-securitygovernancehardeningstableclaude_settingswarn

Claude settings: wildcard Bash permissions

Claude settings permissions allow `Bash(*)` in a shared committed config

Public lane
governance
Category
hardening
Provider
lintai-ai-security
Scope
per-file
Surface
claude_settings
Tier
stable
Severity
warn
Confidence
high
Detection
structural
Remediation
message only
How to read this lane

Shared authority and workflow policy review.

How to read this category

Least-privilege, provenance, or operational hygiene signal.

Activation Model

Preset Activation

These presets explain where this rule appears in the product experience.

Lifecycle

Stable Lifecycle Contract

State

stable

Graduation rationale

Checks shared Claude settings permissions for explicit wildcard `Bash(*)` grants.

Deterministic signal basis

ClaudeSettingsSignals exact string detection for `Bash(*)` inside permissions.allow on parsed Claude settings JSON.

Malicious corpus
claude-settings-bash-wildcard
Benign corpus
claude-settings-bash-specific-safe
structured evidence required remediation reviewed
Canonical note

Structural stable rule intended as a high-precision check with deterministic evidence.

Nearby Signals

Related Rules

Why It Matters

SEC362 flags committed Claude settings files when permissions.allow contains the exact wildcard shell grant Bash(*).

This is useful because:

  • shared .claude/settings.json files are often copied across teams and repos
  • Bash(*) grants unrestricted shell execution instead of a reviewed command allowlist
  • this makes the shared agent policy much broader than most teams intend

What Triggers

This rule applies only to committed Claude settings surfaces:

  • .claude/settings.json
  • claude/settings.json

It triggers when:

  • permissions.allow contains the exact string Bash(*)

It does not trigger on:

  • narrower reviewed patterns such as Bash(git status)
  • fixture-like test/example paths

Examples

Bad:

json
{
  "permissions": {
    "allow": ["Bash(*)", "Read(*)"]
  }
}

Better:

json
{
  "permissions": {
    "allow": ["Bash(git status)", "Bash(npm test:*)", "Read(*)"]
  }
}

Remediation

  • replace Bash(*) with specific reviewed command patterns
  • keep the shared allowlist as narrow as possible for the repo's actual workflows