feat: add root-level tools map with customParameters to AI Config types

[jsonbailey]

Summary

• Adds AITool frozen dataclass with name, type?, parameters?, and custom_parameters? fields
• Adds tools: Optional[Dict[str, AITool]] to AICompletionConfig, AICompletionConfigDefault, AIAgentConfig, and AIAgentConfigDefault
• Adds _parse_tools() helper in client.py that parses the root-level tools map from flag variations
• Updates _completion_config and __evaluate_agent to parse and propagate tools, falling back to default.tools when absent
• Exports AITool from ldai.__init__
• Adds tests/test_tools.py with 6 tests covering tools in completion config, agent config, default fallback, and serialization

Background

The LaunchDarkly backend now includes a root-level tools map (sibling to model) in AI Config flag variations. This is separate from model.parameters.tools[] which is the raw array passable to LLM providers as-is. LLM providers reject unknown properties, so customParameters cannot live inside model.parameters.tools[] — hence the split.

Wire format:

{
  "tools": {
    "web-search-tool": {
      "name": "web-search-tool",
      "type": "function",
      "parameters": { ... },
      "customParameters": { "some-custom-parameter": "some-custom-value" }
    }
  }
}

customParameters is camelCase in the wire format; the Python model uses custom_parameters (snake_case) and serializes back to camelCase in to_dict().

Note

Supersedes PR #106 (devin/1773277027-add-tool-instructions-examples) which had the wrong structure (list instead of map, wrong fields key/version/instructions/examples).

Test plan

• poetry run pytest — all existing tests pass, 6 new tests for tools coverage
• Verify AITool.to_dict() serializes custom_parameters as customParameters

🤖 Generated with Claude Code

---

Note

Medium Risk
Medium risk because it changes the public AI config data model and flag-variation parsing for completion_config/agent_config, which could affect downstream consumers and serialization expectations.

Overview
Adds first-class support for a root-level tools map in AI Config flag variations by introducing the LDTool model (including customParameters wire-format support) and wiring it into AICompletionConfig* and AIAgentConfig*.

Updates LDAIClient evaluation to parse variation['tools'] via _parse_tools() and expose it on returned configs, and exports LDTool from ldai. Adds a focused test suite covering parsing behavior (tools present vs absent) and LDTool.to_dict() camelCase serialization/omissions.

^{Reviewed by Cursor Bugbot for commit 7c522ae. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

There are 2 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit ccc744e. Configure here.}