Claude Guardrails

✓ Policy Experimental Devex Build And Ci

Enforces Claude-specific standards: CI safety flags, structured output, and CLAUDE.md symlink compatibility alongside AGENTS.md.

Add claude to your lunar-config.yml:

uses: github://earthly/lunar-lib/policies/claude@v1.0.5

Get Started View on GitHub

Included Guardrails

This policy includes 3 guardrails that enforce standards for your devex build and ci.

Guardrail

`cli-safe-flags`

Ensures Claude CLI invocations in CI do not use dangerous permission-bypassing flags. Flags like --dangerously-skip-permissions remove safety guardrails that prevent the AI from executing arbitrary code without human approval.

claude ci safety dangerous flags permissions

View Guardrail

Guardrail

`cli-structured-output`

Ensures Claude CLI invocations in CI headless mode use structured JSON output (--output-format json) instead of plain text. JSON output makes AI automation deterministic and parseable.

claude json output structured output ci automation

View Guardrail

Guardrail

`symlinked-aliases`

Verifies that CLAUDE.md exists as a symlink to AGENTS.md. Claude Code does not support the AGENTS.md filename natively, so a symlink is required for compatibility. Reads from ai.native.claude.instruction_file to check symlink status.

claude.md symlink agents.md compatibility

View Guardrail

How Guardrails Fit into Lunar

Lunar guardrails define your engineering standards as code. They evaluate data collected by integrations and produce pass/fail checks with actionable feedback.

Policies support gradual enforcement—from silent scoring to blocking PRs or deployments—letting you roll out standards at your own pace without disrupting existing workflows.

Learn How Lunar Works →

↓ Integrations Gather Data

Collectors extract metadata from code, CI pipelines, tool outputs, and scans

{ } Centralized as JSON

All data merged into each component's unified metadata document

✓ Guardrails Enforce Standards This Policy

Real-time feedback in PRs and AI workflows

Example Evaluated Data

This policy evaluates structured metadata from the Component JSON. Here's an example of the data it checks:

{ } component.json Example Input

{
  "ai": {
    "native": {
      "claude": {
        "instruction_file": {
          "exists": true,
          "path": "CLAUDE.md",
          "lines": 85,
          "bytes": 3200,
          "is_symlink": true,
          "symlink_target": "AGENTS.md"
        },
        "cicd": {
          "cmds": [
            {
              "cmd": "claude -p --output-format json --allowedTools Bash(git*) Read 'review this PR'",
              "tool": "claude",
              "version": "1.0.20",
              "allowed_tools": "Bash(git*) Read"
            }
          ]
        }
      }
    }
  }
}

Required Integrations

This policy evaluates data gathered by one or more of the following integration(s). Make sure to enable them in your lunar-config.yml.

Collector Code Analysis Required

Claude Collector

Provides Claude CI command data and CLAUDE.md instruction file detection

Configuration

Configure this policy in your lunar-config.yml.

Inputs

Input	Required	Default	Description
`dangerous_flags`	Optional	`--dangerously-skip-permissions,--allow-dangerously-skip-permissions`	Comma-separated dangerous flags for Claude CLI

Documentation

View on GitHub

Claude Guardrails

Claude-specific CI safety and conventions guardrails.

Overview

This policy enforces Claude-specific standards: CI safety flags, structured output, and CLAUDE.md symlink compatibility. It validates that Claude CLI invocations do not use dangerous flags, use structured JSON output, and that CLAUDE.md exists as a symlink to AGENTS.md.

Policies

Policy	Severity	Description
`cli-safe-flags`	error	Claude CLI must not use `--dangerously-skip-permissions` or similar flags
`cli-structured-output`	warning	Claude CLI in CI should use `--output-format json`
`symlinked-aliases`	warning	CLAUDE.md must exist as a symlink to AGENTS.md

Required Data

Path	Provided By	Description
`.ai.native.claude.instruction_file`	`claude` collector	CLAUDE.md file detection with symlink status
`.ai.native.claude.cicd.cmds[]`	`claude` collector	Claude CLI invocations captured in CI

Installation

collectors:
  - uses: github://earthly/lunar-lib/collectors/claude@main
    on: ["domain:your-domain"]

policies:
  - uses: github://earthly/lunar-lib/policies/claude@main
    enforcement: report-pr

Examples

Passing

Claude CLI using safe flags and structured output:

{
  "ai": {
    "native": {
      "claude": {
        "cicd": {
          "cmds": [
            {
              "cmd": "claude -p --output-format json --allowedTools Read 'review this PR'",
              "tool": "claude"
            }
          ]
        }
      }
    }
  }
}

Failing

Claude CLI using dangerous permission-bypassing flag:

{
  "ai": {
    "native": {
      "claude": {
        "cicd": {
          "cmds": [
            {
              "cmd": "claude --dangerously-skip-permissions -p 'deploy to prod'",
              "tool": "claude"
            }
          ]
        }
      }
    }
  }
}

Remediation

cli-safe-flags: Remove --dangerously-skip-permissions from Claude CI invocations. Use --allowedTools to grant specific tool access instead.
cli-structured-output: Add --output-format json to Claude CLI invocations in CI.
symlinked-aliases: Create ln -s AGENTS.md CLAUDE.md so Claude Code can find the instruction file.

Open Source

This policy is open source and available on GitHub. Contribute improvements, report issues, or fork it for your own use.

View Repository

Common Use Cases

Explore how individual guardrails work with specific integrations.

Cli Safe Flags + Claude Collector Ensures Claude CLI invocations in CI do not use dangerous...

→

Cli Structured Output + Claude Collector Ensures Claude CLI invocations in CI headless mode use structured JSON...

→

Symlinked Aliases + Claude Collector Verifies that CLAUDE.md exists as a symlink to AGENTS.md. Claude Code does...

→

Ready to Automate Your Standards?

See how Lunar can turn your AGENTS.md, engineering wiki, compliance docs, or postmortem action items into automated guardrails with our 100+ built-in guardrails.

Works with any process

AI agent rules & prompt files

Post-mortem action items

Security & compliance policies

Testing & quality requirements

Automate Now

Paste your AGENTS.md or manual process doc and get guardrails in minutes

Book a Demo