providers.json

{
  "providers": {
    "amp": {
      "name": "AMP",
      "command": "amp",
      "defaultFlags": ["--dangerously-allow-all"],
      "sandboxedFlags": [],
      "description": "Sourcegraph AMP - AI coding agent with full autonomy mode",
      "threadUrlPattern": "https://ampcode.com/threads/${THREAD_ID}",
      "threadIdEnv": "AMP_CURRENT_THREAD_ID",
      "skillsDir": "~/.ralph/skills",
      "documentation": "https://ampcode.com/manual",
      "notes": "Uses --dangerously-allow-all for autonomous execution without confirmation prompts.",
      "sandboxedNotes": "AMP sandboxed mode runs without --dangerously-allow-all, requiring user approval for each action. Check ampcode.com/manual for allowlist options when available.",
      "supportsParallel": true,
      "parallelNotes": "Fully supports parallel execution. Each worker uses separate threads. Recommended for independent stories."
    },
    "claude-code": {
      "name": "Claude Code",
      "command": "claude",
      "defaultFlags": ["--dangerously-skip-permissions"],
      "sandboxedFlags": ["--allowedTools", "Read,Edit,Write,Bash(npm run *),Bash(pnpm run *),Bash(yarn *),Bash(cargo *),Bash(go *),Bash(make *),Bash(git status),Bash(git diff*),Bash(git log*),Bash(git show*),Bash(git branch*),Bash(cat *),Bash(ls*),Bash(pwd),Bash(grep *),Bash(find *),Bash(jq *)"],
      "description": "Anthropic Claude Code - Claude-powered coding assistant",
      "threadUrlPattern": "",
      "threadIdEnv": "",
      "skillsDir": "~/.ralph/skills",
      "documentation": "https://docs.anthropic.com/en/docs/claude-code",
      "notes": "Uses --dangerously-skip-permissions for autonomous execution. Alternative: -p flag for print-only mode during testing.",
      "sandboxedNotes": "Uses --allowedTools to restrict to safe operations (lint, test, build, read-only git). See config/permissions.json for full allowlist.",
      "supportsParallel": true,
      "parallelNotes": "Supports parallel execution with separate sessions. Each worker runs in isolation. Monitor for rate limits with multiple instances."
    },
    "antigravity": {
      "name": "Antigravity",
      "command": "antigravity",
      "defaultFlags": ["--auto-approve"],
      "sandboxedFlags": [],
      "description": "Google DeepMind Antigravity - Advanced agentic coding assistant",
      "threadUrlPattern": "",
      "threadIdEnv": "",
      "skillsDir": "~/.ralph/skills",
      "documentation": "",
      "notes": "Assumes --auto-approve flag for autonomous execution. Adjust if CLI differs.",
      "sandboxedNotes": "Sandboxed mode not yet documented. Runs without --auto-approve, requiring manual approval.",
      "supportsParallel": false,
      "parallelNotes": "Parallel execution not yet verified. CLI behavior with multiple instances unknown. Use sequential mode."
    },
    "codex": {
      "name": "CODEX",
      "command": "codex",
      "defaultFlags": ["--approval-mode", "full-auto"],
      "sandboxedFlags": ["--approval-mode", "suggest"],
      "description": "OpenAI CODEX CLI - GPT-powered coding agent",
      "threadUrlPattern": "",
      "threadIdEnv": "",
      "skillsDir": "~/.ralph/skills",
      "documentation": "https://github.com/openai/codex",
      "notes": "Uses --approval-mode full-auto for autonomous execution without prompts.",
      "sandboxedNotes": "Uses --approval-mode suggest to require approval for file writes and commands.",
      "supportsParallel": true,
      "parallelNotes": "Supports parallel execution. Be aware of API rate limits when running multiple workers."
    }
  },
  "default": "amp",
  "_meta": {
    "version": "1.1.0",
    "description": "Ralph AI provider configuration. Add new providers by following the schema above.",
    "flagRationale": {
      "amp": "Official AMP documentation recommends --dangerously-allow-all for automated pipelines.",
      "claude-code": "Claude Code CLI uses --dangerously-skip-permissions to bypass file/command confirmation prompts.",
      "antigravity": "Assumed flag based on common patterns. Verify with actual CLI documentation.",
      "codex": "OpenAI Codex CLI uses --approval-mode full-auto for fully autonomous operation."
    },
    "sandboxedModeInfo": {
      "description": "Sandboxed mode (RALPH_SANDBOXED=1) uses sandboxedFlags instead of defaultFlags for safer operation.",
      "permissionsFile": "config/permissions.json contains the detailed allowlist/blocklist rules.",
      "recommendation": "Use sandboxed mode for untrusted codebases or when testing new PRDs."
    }
  }
}