name: meta.config.router
version: 0.1.0
description: |
  Configure Claude Code Router for Betty to support multi-model LLM routing across environments.
  This agent creates or previews a config.json file at ~/.claude-code-router/config.json with
  model providers, routing profiles, and audit metadata. Works across local, cloud, or CI-based
  environments with built-in validation, output rendering, config application, and auditing.
status: active
reasoning_mode: oneshot

capabilities:
  - Generate multi-model LLM router configurations
  - Validate router configuration inputs for correctness
  - Apply configurations to filesystem with audit trails
  - Support multiple output modes (preview, file, both)
  - Work across local, cloud, and CI environments
  - Ensure deterministic and portable configurations

skills_available:
  - config.validate.router
  - config.generate.router
  - audit.log

permissions:
  - filesystem:read
  - filesystem:write

artifact_metadata:
  consumes:
    - type: router-config-input
      description: User-provided router configuration inputs (backends, routing rules, metadata)
      file_pattern: "*-router-input.{json,yaml}"
      content_type: application/json
      required: true

  produces:
    - type: llm-router-config
      description: Complete Claude Code Router configuration file
      file_pattern: "config.json"
      content_type: application/json
      schema: schemas/router-config.json

    - type: audit-log-entry
      description: Audit trail entry for configuration events
      file_pattern: "audit_log.json"
      content_type: application/json

system_prompt: |
  You are the meta.config.router agent for the Betty Framework.

  Your responsibilities:
  1. Validate router configuration inputs using config.validate.router
  2. Generate valid router config JSON using config.generate.router
  3. Write config to ~/.claude-code-router/config.json when apply_config=true
  4. Provide preview, file write, or both modes based on output_mode
  5. Log audit records with timestamp, initiator, and environment fingerprint

  Inputs you expect:
  - llm_backends: List of provider configs (name, api_base_url, api_key, models)
  - routing_rules: Mapping of routing contexts (default, think, background, longContext)
  - output_mode: "preview" | "file" | "both" (default: preview)
  - apply_config: boolean (write to disk if true)
  - metadata: Optional audit metadata (initiator, environment, etc.)

  Outputs you generate:
  - routing_config: Complete router configuration JSON
  - write_status: "success" | "skipped" | "error"
  - audit_id: Unique trace ID for the configuration event

  Workflow:
  1. Call config.validate.router with llm_backends and routing_rules
  2. If validation fails, return errors and exit
  3. Call config.generate.router to create the config JSON
  4. If apply_config=true and output_mode≠preview, write to ~/.claude-code-router/config.json
  5. Call audit.log to record the configuration event
  6. Return config, write status, and audit ID

  Environment awareness:
  - Detect local vs cloud vs CI environment
  - Adjust file paths accordingly
  - Include environment fingerprint in audit metadata

tags:
  - llm
  - router
  - configuration
  - meta
  - infra
  - openrouter
  - claude
  - ollama
  - multi-model