fix(agent base): fixed many iterations and tool calls

[ChrisCoder9000]

Summary by CodeRabbit

• Chores

  • Updated version to 2.9.1-dev

• New Features

  • Enhanced agent system with improved tool orchestration and structured response handling
  • Added advanced language model configuration options for optimized processing

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 5ef7de0d-677d-467b-ac86-1a3466a271e7

📥 Commits

Reviewing files that changed from the base of the PR and between 1738711 and e130dff.

⛔ Files ignored due to path filters (1)

• uv.lock is excluded by !**/*.lock

📒 Files selected for processing (12)

• README.md
• pyproject.toml
• src/core/agents/core/__init__.py
• src/core/agents/core/agent_base.py
• src/core/agents/core/invoke_loop.py
• src/core/agents/core/parsing.py
• src/core/agents/core/prompts.py
• src/core/agents/core/schema_utils.py
• src/core/agents/core/structured_output.py
• src/core/agents/core/types.py
• src/core/agents/scout_agent.py
• src/lib/llm/client_small.py

---

📝 Walkthrough

Walkthrough

This PR refactors the agent core infrastructure by extracting complex orchestration logic from agent_base.py into five new specialized modules (invoke_loop, parsing, prompts, schema_utils, structured_output). The main invoke method now delegates to an external run_invoke_loop function, reducing agent_base.py from 1244 to 12 lines of implementation logic. Version bumped to 2.9.1-dev across configuration files.

Changes

Cohort / File(s)	Summary
Version Bumps README.md, pyproject.toml	Updated version badge and project version from 2.9.0-dev to 2.9.1-dev.
Core Agent Refactoring src/core/agents/core/agent_base.py	Replaced 1200+ lines of internal orchestration logic (message normalization, tool invocation, parsing) with single delegation call to run_invoke_loop. Updated output_schema typing and simplified methods like _get_effective_output_schema and invoke to delegate to external functions. Removed legacy tool-call parsing and structured response validation.
New Invoke Loop Module src/core/agents/core/invoke_loop.py	New 588-line module implementing run_invoke_loop(agent, messages, config) with iterative agent-model interaction, tool call extraction and invocation, thought signature handling, schema-based output parsing, and error retry logic for unknown finish reasons.
New Parsing Utilities src/core/agents/core/parsing.py	New 509-line module providing 12 functions for message normalization, tool name/call extraction, thought signature extraction, invoke response normalization, and JSON repair utilities. Handles varied response shapes (dicts, objects) across different LLM providers.
New Schema & Output Utilities src/core/agents/core/schema_utils.py, src/core/agents/core/structured_output.py	Schema utils (146 lines): functions for effective schema resolution, JSON schema derivation, list-field validation, and schema flattening. Structured output (145 lines): JSON parsing with repair, Pydantic model validation, and fallback handling via list-field detection.
New Prompt Building Module src/core/agents/core/prompts.py	New 67-line module with two functions: format_tool_description (renders tool metadata) and build_system_internal_prompt (constructs system prompts with tools, output schema, and optional thinking instructions).
New Type Definitions src/core/agents/core/types.py	New 17-line module defining AgentMessage, AgentOutput, and MessagesDict TypedDict structures for consistent agent messaging and response shapes.
Export API Updates src/core/agents/core/__init__.py	Added explicit imports and exports of AgentBase and parse_structured_from_messages to update the public API surface.
Minor Updates src/core/agents/scout_agent.py	Last Modified header date change only; no functional changes.
LLM Client Configuration src/lib/llm/client_small.py	Added thinking_budget: 0 to ChatVertexAI initialization. Introduced dynamic config_kw for JSON generation with conditional ThinkingConfig inclusion and refactored GenerateContentConfig to use dynamic kwargs.

Sequence Diagram(s)

sequenceDiagram
    actor User
    participant Agent as AgentBase
    participant InvokeLoop as run_invoke_loop()
    participant Model as LLM Model
    participant Tools as Tool Executor
    participant Parser as Parsing Utils
    participant SchemaUtils as Schema Utils

    User->>Agent: invoke(messages, config)
    Agent->>InvokeLoop: run_invoke_loop(self, messages, config)
    InvokeLoop->>InvokeLoop: Build system prompt + append messages
    loop Iterative Invocation
        InvokeLoop->>Model: Invoke with messages
        Model-->>InvokeLoop: Response (may include tool_calls)
        InvokeLoop->>Parser: Extract tool calls from response
        Parser-->>InvokeLoop: tool_name, tool_input
        alt Tool Call Detected
            InvokeLoop->>Tools: Execute tool(tool_name, input)
            Tools-->>InvokeLoop: tool_result
            InvokeLoop->>InvokeLoop: Append tool result to messages
        else No Tool Call
            InvokeLoop->>InvokeLoop: Check finish reason
            alt UNKNOWN Finish Reason
                InvokeLoop->>InvokeLoop: Retry with continuation prompt
            else Finished
                InvokeLoop->>InvokeLoop: Exit loop
            end
        end
    end
    InvokeLoop->>Parser: parse_structured_from_messages()
    Parser->>SchemaUtils: get_effective_output_schema()
    SchemaUtils-->>Parser: effective_schema
    Parser->>Parser: Parse structured response against schema
    Parser-->>InvokeLoop: structured_response
    InvokeLoop-->>Agent: {messages, structured_response}
    Agent-->>User: AgentOutput

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~75 minutes

Possibly related PRs

• PR #23: Directly conflicts with and modifies the same agent core infrastructure (agent_base.py and package exports), representing concurrent or competing refactoring efforts on the core agent architecture.

Poem

🐰 The rabbit hops through code so grand,
Breaking orchards into smaller lands,
Loops and parsers, schemas bright,
From chaos springs a cleaner light!
Where once lived giants, now small friends,
Orchestrate where logic bends. 🌱

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch development

Warning

There were issues while running some tools. Please review the errors and either fix the tool's configuration or disable the tool if it's a critical failure.

🔧 Pylint (4.0.5)

src/core/agents/core/invoke_loop.py

************* Module .pylintrc
.pylintrc:1:0: F0011: error while parsing the configuration: Source contains parsing errors: '.pylintrc'
[line 18]: 'C0103' (config-parse-error)
[
{
"type": "convention",
"module": "src.core.agents.core.invoke_loop",
"obj": "",
"line": 179,
"column": 0,
"endLine": null,
"endColumn": null,
"path": "src/core/agents/core/invoke_loop.py",
"symbol": "line-too-long",
"message": "Line too long (102/100)",
"message-id": "C0301"
},
{
"type": "convention",
"module": "src.core.agents.core.invoke_loop",
"obj": "",
"line": 195,
"column": 0,
"endLine": null,
"endColumn": null,
"path": "src/core/agents/core/invoke_loop.py",
"symbol": "line-too-long",
"message": "Line too long (147/100)",
"message-id": "C0301"
},
{
"type": "convention",
"module": "src.core.agent

... [truncated 12725 characters] ...

"W0212"
},
{
"type": "refactor",
"module": "src.core.agents.core.invoke_loop",
"obj": "run_invoke_loop",
"line": 20,
"column": 0,
"endLine": 20,
"endColumn": 19,
"path": "src/core/agents/core/invoke_loop.py",
"symbol": "too-many-branches",
"message": "Too many branches (83/12)",
"message-id": "R0912"
},
{
"type": "refactor",
"module": "src.core.agents.core.invoke_loop",
"obj": "run_invoke_loop",
"line": 20,
"column": 0,
"endLine": 20,
"endColumn": 19,
"path": "src/core/agents/core/invoke_loop.py",
"symbol": "too-many-statements",
"message": "Too many statements (269/50)",
"message-id": "R0915"
}
]

src/core/agents/core/agent_base.py

************* Module .pylintrc
.pylintrc:1:0: F0011: error while parsing the configuration: Source contains parsing errors: '.pylintrc'
[line 18]: 'C0103' (config-parse-error)
[
{
"type": "convention",
"module": "src.core.agents.core.agent_base",
"obj": "",
"line": 1,
"column": 0,
"endLine": null,
"endColumn": null,
"path": "src/core/agents/core/agent_base.py",
"symbol": "missing-module-docstring",
"message": "Missing module docstring",
"message-id": "C0114"
},
{
"type": "error",
"module": "src.core.agents.core.agent_base",
"obj": "",
"line": 4,
"column": 0,
"endLine": 4,
"endColumn": 16,
"path": "src/core/agents/core/agent_base.py",
"symbol": "import-error",
"message": "Unable to import 'langsmith'",
"message-id": "E0401"
},
{
"type": "error",
"module": "src.core.agents.core.

... [truncated 3105 characters] ...

},
{
    "type": "refactor",
    "module": "src.core.agents.core.agent_base",
    "obj": "AgentBase",
    "line": 21,
    "column": 0,
    "endLine": 21,
    "endColumn": 15,
    "path": "src/core/agents/core/agent_base.py",
    "symbol": "too-few-public-methods",
    "message": "Too few public methods (1/2)",
    "message-id": "R0903"
},
{
    "type": "warning",
    "module": "src.core.agents.core.agent_base",
    "obj": "",
    "line": 10,
    "column": 0,
    "endLine": 10,
    "endColumn": 51,
    "path": "src/core/agents/core/agent_base.py",
    "symbol": "unused-import",
    "message": "Unused parse_structured_from_messages imported from parsing",
    "message-id": "W0611"
}

]

src/core/agents/core/prompts.py

************* Module .pylintrc
.pylintrc:1:0: F0011: error while parsing the configuration: Source contains parsing errors: '.pylintrc'
[line 18]: 'C0103' (config-parse-error)
[
{
"type": "convention",
"module": "src.core.agents.core.prompts",
"obj": "",
"line": 1,
"column": 0,
"endLine": null,
"endColumn": null,
"path": "src/core/agents/core/prompts.py",
"symbol": "missing-module-docstring",
"message": "Missing module docstring",
"message-id": "C0114"
},
{
"type": "error",
"module": "src.core.agents.core.prompts",
"obj": "",
"line": 4,
"column": 0,
"endLine": 4,
"endColumn": 36,
"path": "src/core/agents/core/prompts.py",
"symbol": "import-error",
"message": "Unable to import 'langchain.tools'",
"message-id": "E0401"
},
{
"type": "convention",
"module": "src.core.agents.core.p

... [truncated 321 characters] ...

116"
},
{
"type": "convention",
"module": "src.core.agents.core.prompts",
"obj": "build_system_internal_prompt",
"line": 23,
"column": 0,
"endLine": 23,
"endColumn": 32,
"path": "src/core/agents/core/prompts.py",
"symbol": "missing-function-docstring",
"message": "Missing function or method docstring",
"message-id": "C0116"
},
{
"type": "warning",
"module": "src.core.agents.core.prompts",
"obj": "",
"line": 2,
"column": 0,
"endLine": 2,
"endColumn": 32,
"path": "src/core/agents/core/prompts.py",
"symbol": "unused-import",
"message": "Unused Optional imported from typing",
"message-id": "W0611"
}
]

• 6 others

Tip

Try Coding Plans. Let us write the prompt for your AI agent so you can ship faster (with fewer bugs).
Share your feedback on Discord.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}