fix: wire named agents + 7 tests for managed agents

[kienbui1995]

Named agents wired to subagent dispatch

{"task": "review auth", "agent_name": "reviewer"}

Looks up agents/reviewer.md → uses its model, tools, instructions.

7 new tests (183→190)

• SubagentSpawner: max_concurrent, background poll/list, shared context
• ManagedAgentConfig: TOML parse, validation, defaults

Addresses CodeRabbit gap: named agents were visible in prompt but not routed.

Summary by CodeRabbit

• New Features

  • Subagents can now reference predefined named agents, automatically inheriting their configured models, tools, and instructions.
  • Configuration validation for managed agents now checks budget and execution limits.

• Tests

  • Added test coverage for agent configuration parsing and subagent execution behavior.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 8615d5e2-f902-4885-9cf8-9d71d333434e

📥 Commits

Reviewing files that changed from the base of the PR and between 7f5aa3b and c513c06.

📒 Files selected for processing (4)

• mc/crates/mc-config/src/types.rs
• mc/crates/mc-core/src/runtime.rs
• mc/crates/mc-core/src/subagent.rs
• mc/crates/mc-tools/src/spec.rs

---

📝 Walkthrough

Walkthrough

The changes introduce agent definition support to ConversationRuntime, enabling the subagent tool to look up named agent configurations and derive model overrides and allowed tools dynamically. Configuration validation tests verify managed agents parsing and defaults. Tool schema updated to accept optional agent_name parameter. Additional unit tests cover agent lookup, subagent spawning, and context management.

Changes

Cohort / File(s)	Summary
Runtime Agent Support mc/crates/mc-core/src/runtime.rs	Added agents field to ConversationRuntime with set_agents() setter. Modified subagent tool dispatcher to read agent_name input, look up agent definition, derive model_override and allowed_tools, and prepend agent instructions to prompt.
Tool Schema Updates mc/crates/mc-tools/src/spec.rs	Added optional agent_name string field to subagent tool's input schema to reference named agents from configuration.
Configuration Testing mc/crates/mc-config/src/types.rs	Added unit tests validating managed_agents TOML parsing, RuntimeConfig validation warnings for invalid values, and default initialization when no managed_agents block is provided.
Subagent Testing mc/crates/mc-core/src/subagent.rs	Added unit tests for SubagentSpawner concurrent configuration, missing agent ID handling, background result listing, and SharedContext key/value round-trip persistence.

Sequence Diagram

sequenceDiagram
    participant Agent as Conversation Agent
    participant Runtime as ConversationRuntime
    participant Agents as Agents Registry
    participant SubTool as Subagent Tool Handler
    participant Task as Task Executor
    
    Agent->>Runtime: dispatch_tool("subagent", {agent_name: "foo"})
    Runtime->>Runtime: lookup agent_name in self.agents
    Runtime->>Agents: AgentDef found
    Agents-->>Runtime: return AgentDef (model, tools, instructions)
    Runtime->>SubTool: construct sub_prompt with agent instructions
    SubTool->>Task: run_task(model_override: agent.model)
    Task-->>SubTool: result
    SubTool-->>Runtime: tool result

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• feat: named agents, audit timestamps, --trace logging #32 — Introduces AgentDef structure and agent discovery mechanisms that directly align with the agents registry implementation added here.
• feat: managed agents — manager-executor pattern #34 — Modifies subagent flow with agent-model routing and new subagent input fields, directly complementing the agent lookup and dispatch logic.
• feat: address critical gaps — model routing, CI mode, per-tool permissions #27 — Adds explicit model_override parameter to dispatch/run_task behavior, which this PR now leverages when deriving model from agent definition.

Poem

A rabbit hops through agent fields so neat,
Subagents find their instructions sweet,
Model overrides dance, tools align,
Config tests ensure all schemas shine! 🐰✨

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/managed-agents-gaps

---

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