Skills

Maybe Don’t ships AI agent skills — pre-written instructions embedded in the binary that teach AI coding agents how to use the gateway’s features. Export them and give them to your agent.

What Are Skills?

Skills are prompt templates. They contain schema references, examples, and instructions that teach an AI agent how to write CEL policies, AI policies, test cases, or route CLI commands through the proxy. Instead of copy-pasting documentation into a chat, you export a skill and the agent has everything it needs.

Available Skills

Skill	Description
`cli`	Teaches an agent to route CLI commands through the proxy
`cel-policy`	Guide for authoring CEL deterministic policy rules
`ai-policy`	Guide for authoring AI LLM-powered policy rules
`test-case`	Guide for writing policy test cases and configuring test suites

Listing Skills

maybe-dont skill list

Exporting Skills

Use skill view to output a skill in a format your agent understands:

# Claude Code
maybe-dont skill view cel-policy --format claude > .claude/skills/cel-policy.md

# Cursor
maybe-dont skill view cel-policy --format cursor >> .cursorrules

# GitHub Copilot
maybe-dont skill view cel-policy --format copilot > .github/copilot-instructions.md

# Generic (any agent)
maybe-dont skill view cel-policy --format generic > instructions.md

Output Formats

Format	Target	Description
`claude`	Claude Code	Full skill with markdown sections
`cursor`	Cursor IDE	Concise `.cursorrules` format
`copilot`	GitHub Copilot	Copilot instructions markdown
`generic`	Any agent	General-purpose markdown

Workflow: Writing Policies and Tests

Here’s how to use skills end-to-end:

Export the policy skill to your agent:

maybe-dont skill view ai-policy --format claude > .claude/skills/ai-policy.md

Describe the policy you want in natural language — the agent uses the skill’s schema reference and examples to produce valid YAML

Export the test-case skill:

maybe-dont skill view test-case --format claude > .claude/skills/test-case.md

Ask the agent to generate test cases for the policy it just wrote

Run the tests to validate:

maybe-dont test policies --suite-dir ./suite

Schema Reference

Quick-reference schemas showing every field. The embedded skills include richer context — CEL variables, available functions, prompt patterns, and worked examples — that helps AI agents produce correct YAML. Use maybe-dont skill view <name> to see the full version, or export it to your agent with --format.

CEL Request Rule

name: ""                   # Unique rule identifier (kebab-case)
description: ""            # Human-readable description
enabled: true              # true or false
mcp_expression: ""         # CEL expression for MCP tool calls
cli_expression: ""         # CEL expression for CLI commands
action: deny               # deny or allow
message: ""                # Message returned when rule triggers
mode: ""                   # audit_only or enforce (empty = enforce)

AI Request Rule

name: ""                   # Unique rule identifier (kebab-case)
description: ""            # Human-readable description
enabled: true              # true or false
action: deny               # deny or allow
mode: ""                   # audit_only or enforce (empty = enforce)
prompt: ""                 # Natural language policy instruction

Test Case

case_id: ""                # Unique ID across the suite (kebab-case)
title: ""                  # Human-readable test description
tags: []                   # For --tags/--exclude-tags filtering
notes: []                  # Documentation strings
phase: request             # request, response, or both
engine: both               # cel, ai, or both
request:
  tool_name: ""            # MCP tool name (MCP tests)
  arguments: {}            # Tool arguments
  command: ""              # CLI command name (CLI tests)
  cli_arguments: []        # CLI command arguments
expectations:
  decision: deny           # allow, deny, or redact
  policies:
    - policy_name: ""      # Expected triggering policy
      decision: deny       # Expected policy decision

Configuration Native Tools