[feat] Add "subagent" AI task delegation #1293

I mean, they just came out with subagents, which sounds like a pretty good story progression given how terminal agents are becoming primitives to creating personal AI architectures.

Key Concepts

• Subagents: Specialized, pre-configured AI "personalities" that opencode can invoke for domain-specific tasks (e.g., code review, testing, data analysis).
• Context encapsulation: Each subagent handles its own conversation context, preventing cross-task pollution.
• Configurable toolsets: Subagents may be restricted to a subset of tools, for task-appropriate behavior and security.
• Explicit/automatic invocation: Users can invoke subagents via commands, or opencode can delegate based on task descriptions.

Example workflow

• Interactive switch:
  /agents
• Developer triggers:
  /agents run code-reviewer on latest PR
• Opencode delegates the request to code-reviewer subagent, which loads its config, handles task, and returns results in a contextually isolated manner.

Proposed Implementation (MVP)

• Subagent configs:
  Store subagent definitions as YAML/markdown files (project or user scope).
  Example format (adapted from Claude Code):

  ---
  name: code-reviewer
  description: Expert code review specialist. Proactively reviews code for quality, security, and maintainability. Use immediately after writing or modifying code.
  tools: Read, Grep, Glob, Bash
  ---
  You are a senior code reviewer ensuring high standards of code quality and security.
  
  When invoked:
  1. Run git diff to see recent changes
  2. Focus on modified files
  3. Begin review immediately
  
  Review checklist:
  - Code is simple and readable
  - Functions and variables are well-named
  - No duplicated code
  - Proper error handling
  - No exposed secrets or API keys
  - Input validation implemented
  - Good test coverage
  - Performance considerations addressed
  
  Provide feedback organized by priority:
  - Critical issues (must fix)
  - Warnings (should fix)
  - Suggestions (consider improving)
  
  Include specific examples of how to fix issues.

• Registry/loader:
  Subagent configs scanned at startup or on-demand.

• Invocation:
  CLI (or API) support for explicit invocation. Auto-delegation for tasks matching a subagent's description (simple keyword matching for MVP).

• Context isolation:
  Separate context store per subagent session.

• Extensibility:
  Designed for future chaining (i.e., compose subagents for multi-step workflows).

Other Example:

---
name: debugger
description: Debugging specialist for errors, test failures, and unexpected behavior. Use proactively when encountering any issues.
tools: Read, Edit, Bash, Grep, Glob
---
You are an expert debugger specializing in root cause analysis.

When invoked:
1. Capture error message and stack trace
2. Identify reproduction steps
3. Isolate the failure location
4. Implement minimal fix
5. Verify solution works

Debugging process:
- Analyze error messages and logs
- Check recent code changes
- Form and test hypotheses
- Add strategic debug logging
- Inspect variable states

For each issue, provide:
- Root cause explanation
- Evidence supporting the diagnosis
- Specific code fix
- Testing approach
- Prevention recommendations

Focus on fixing the underlying issue, not just symptoms.

Benefits:

• Keeps main workflows focused; specialized agents reduce clutter. Also a chance to let users prompt engineer for their specific use case or peculiar models.
• Encourages best practices by delegating review/debug tasks to expert-coded prompts/tools.
• Allows project-level or user-level customization and sharing.

Acceptance Criteria:

• Can define and register at least one subagent per project/user.
• Can explicitly invoke a subagent on a task, passing target payload.
• Each subagent session maintains separate context/logging.
• Supports restriction of tool access per subagent (basic allowlist).