Skip to content

configuration.md

Latest commit

 

History

History
71 lines (55 loc) · 1.7 KB

configuration.md

File metadata and controls

71 lines (55 loc) · 1.7 KB
Raw

Configuration

re-agent is configured via re-agent.yaml, environment variables, and CLI flags.

Priority Order

CLI flags > Environment variables > YAML config > Defaults

Environment Variables

Variable Maps to
RE_AGENT_LLM_PROVIDER llm.provider
RE_AGENT_LLM_API_KEY llm.api_key
RE_AGENT_LLM_MODEL llm.model
RE_AGENT_LLM_BASE_URL llm.base_url
RE_AGENT_BACKEND_CLI_PATH backend.cli_path
RE_AGENT_BACKEND_TIMEOUT backend.timeout_s

LLM Config

llm:
  provider: "claude"        # claude | openai | openai-compat | codex
  model: "claude-sonnet-4-5-20250929"
  api_key: null
  base_url: null
  max_tokens: 4096
  temperature: 0.0
  timeout_s: 1800

Notes:

  • claude uses the Anthropic SDK and typically reads ANTHROPIC_API_KEY
  • openai and openai-compat use the OpenAI-compatible chat completions provider and typically read OPENAI_API_KEY
  • codex uses the local codex CLI and ChatGPT login credentials instead of an API key

Project Profile

The project_profile section makes re-agent work across different RE projects:

project_profile:
  hook_patterns:
    - 'RH_ScopedInstall\s*\(\s*(\w+)\s*,\s*(0x[0-9A-Fa-f]+)'
  stub_markers: ["NOTSA_UNREACHABLE"]
  stub_call_prefix: "plugin::Call"
  source_root: "./source/game_sa"
  source_extensions: [".cpp", ".h", ".hpp"]

Parity Config

parity:
  enabled: true
  call_count_warn_diff: 3
  inline_wrapper_autoskip: false

Orchestrator Config

orchestrator:
  max_review_rounds: 4
  max_functions_per_class: 10
  objective_verifier_enabled: true
  objective_call_count_tolerance: 3
  objective_control_flow_tolerance: 2