Files
amc/docs/claude-jsonl-reference/04-subagent-teams.md
teernisse 781e74cda2 docs(jsonl): add comprehensive Claude JSONL session log reference
Create authoritative documentation suite for Claude Code JSONL session
log processing, synthesized from codebase analysis, official Anthropic
documentation, and community tooling research.

Documentation structure (docs/claude-jsonl-reference/):

01-format-specification.md (214 lines):
- Complete message envelope structure with all fields
- Content block types (text, thinking, tool_use, tool_result)
- Usage object for token reporting
- Model identifiers and version history
- Conversation DAG structure via parentUuid

02-message-types.md (346 lines):
- Every message type with concrete JSON examples
- User messages (string content vs array for tool results)
- Assistant messages with all content block variants
- Progress events (hooks, bash, MCP)
- System, summary, and file-history-snapshot types
- Codex format differences (response_item, function_call)

03-tool-lifecycle.md (341 lines):
- Complete tool invocation to result flow
- Hook input/output formats (PreToolUse, PostToolUse)
- Parallel tool call handling
- Tool-to-result pairing algorithm
- Missing result edge cases
- Codex tool format differences

04-subagent-teams.md (363 lines):
- Task tool invocation and input fields
- Subagent transcript locations and format
- Team coordination (TeamCreate, SendMessage)
- Hook events (SubagentStart, SubagentStop)
- AMC spawn tracking with pending spawn registry
- Worktree isolation for subagents

05-edge-cases.md (475 lines):
- Parsing edge cases (invalid JSON, type ambiguity)
- Type coercion gotchas (bool vs int in Python)
- Session state edge cases (orphans, dead detection)
- Tool call edge cases (missing results, parallel ordering)
- Codex-specific quirks (content injection, buffering)
- File system safety (path traversal, permissions)
- Cache invalidation strategies

06-quick-reference.md (238 lines):
- File locations cheat sheet
- jq recipes for common queries
- Python parsing snippets
- Common gotchas table
- Useful constants
- Debugging commands

Also adds CLAUDE.md at project root linking to documentation and
providing project overview for agents working on AMC.

Sources include Claude Code hooks.md, headless.md, Anthropic Messages
API reference, and community tools (claude-code-log, claude-JSONL-browser).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-02-28 00:48:55 -05:00

7.7 KiB

Subagent and Team Message Formats

Documentation for spawned agents, team coordination, and inter-agent messaging.

Subagent Overview

Subagents are spawned via the Task tool and run in separate processes with their own transcripts.

Spawn Relationship

Main Session (session-uuid.jsonl)
├── User message
├── Assistant: Task tool_use
├── [Subagent executes in separate process]
├── User message: tool_result with subagent output
└── ...

Subagent Session (session-uuid/subagents/agent-id.jsonl)
├── Subagent receives prompt
├── Subagent works (tool calls, etc.)
└── Subagent returns result

Task Tool Invocation

Spawning a subagent:

{
  "type": "assistant",
  "message": {
    "content": [
      {
        "type": "tool_use",
        "id": "toolu_01TaskSpawn",
        "name": "Task",
        "input": {
          "description": "Research auth patterns",
          "prompt": "Investigate authentication implementations in the codebase. Focus on JWT handling and session management.",
          "subagent_type": "Explore",
          "run_in_background": false
        }
      }
    ]
  }
}

Task Tool Input Fields

Field Type Required Description
description string Yes Short (3-5 word) description
prompt string Yes Full task instructions
subagent_type string Yes Agent type (Explore, Plan, etc.)
run_in_background boolean No Run async without waiting
model string No Override model (sonnet, opus, haiku)
isolation string No "worktree" for isolated git copy
team_name string No Team to join
name string No Agent display name

Subagent Types

Type Tools Available Use Case
Explore Read-only tools Research, search, analyze
Plan Read-only tools Design implementation plans
general-purpose All tools Full implementation
claude-code-guide Docs tools Answer Claude Code questions
Custom agents Defined in .claude/agents/ Project-specific

Subagent Transcript Location

~/.claude/projects/<project-hash>/<session-id>/subagents/agent-<agent-id>.jsonl

Subagent Message Format

Subagent transcripts have additional context fields:

{
  "parentUuid": null,
  "isSidechain": true,
  "userType": "external",
  "cwd": "/Users/dev/myproject",
  "sessionId": "subagent-session-uuid",
  "version": "2.1.20",
  "gitBranch": "main",
  "agentId": "a3fecd5",
  "type": "user",
  "message": {
    "role": "user",
    "content": "Investigate authentication implementations..."
  },
  "uuid": "msg-001",
  "timestamp": "2026-02-27T10:00:00.000Z"
}

Key Differences from Main Session

Field Main Session Subagent Session
isSidechain false true
agentId absent present
sessionId main session UUID subagent session UUID

Task Result

When subagent completes, result returns to main session:

{
  "type": "user",
  "message": {
    "content": [
      {
        "type": "tool_result",
        "tool_use_id": "toolu_01TaskSpawn",
        "content": "## Authentication Research Findings\n\n### JWT Implementation\n- Located in src/auth/jwt.ts\n- Uses RS256 algorithm\n...\n\nagentId: a3fecd5 (for resuming)"
      }
    ]
  }
}

Note: Result includes agentId for potential resumption.

Background Tasks

For run_in_background: true:

{
  "type": "tool_use",
  "name": "Task",
  "input": {
    "prompt": "Run comprehensive test suite",
    "subagent_type": "general-purpose",
    "run_in_background": true
  }
}

Immediate result with task ID:

{
  "type": "tool_result",
  "content": "Background task started.\nTask ID: task-abc123\nUse TaskOutput tool to check status."
}

Team Coordination

Teams enable multiple agents working together.

Team Creation (TeamCreate Tool)

{
  "type": "tool_use",
  "name": "TeamCreate",
  "input": {
    "team_name": "auth-refactor",
    "description": "Refactoring authentication system"
  }
}

Team Config File

Created at ~/.claude/teams/<team-name>/config.json:

{
  "team_name": "auth-refactor",
  "description": "Refactoring authentication system",
  "created_at": "2026-02-27T10:00:00.000Z",
  "members": [
    {
      "name": "team-lead",
      "agentId": "agent-lead-123",
      "agentType": "general-purpose"
    },
    {
      "name": "researcher",
      "agentId": "agent-research-456",
      "agentType": "Explore"
    }
  ]
}

Spawning Team Members

{
  "type": "tool_use",
  "name": "Task",
  "input": {
    "prompt": "Research existing auth implementations",
    "subagent_type": "Explore",
    "team_name": "auth-refactor",
    "name": "researcher"
  }
}

Inter-Agent Messaging (SendMessage)

Direct Message

{
  "type": "tool_use",
  "name": "SendMessage",
  "input": {
    "type": "message",
    "recipient": "researcher",
    "content": "Please focus on JWT refresh token handling",
    "summary": "JWT refresh priority"
  }
}

Broadcast to Team

{
  "type": "tool_use",
  "name": "SendMessage",
  "input": {
    "type": "broadcast",
    "content": "Critical: Found security vulnerability in token validation",
    "summary": "Security alert"
  }
}

Shutdown Request

{
  "type": "tool_use",
  "name": "SendMessage",
  "input": {
    "type": "shutdown_request",
    "recipient": "researcher",
    "content": "Task complete, please wrap up"
  }
}

Shutdown Response

{
  "type": "tool_use",
  "name": "SendMessage",
  "input": {
    "type": "shutdown_response",
    "request_id": "req-abc123",
    "approve": true
  }
}

Hook Events for Subagents

SubagentStart Hook Input

{
  "session_id": "main-session-uuid",
  "transcript_path": "/path/to/main/session.jsonl",
  "hook_event_name": "SubagentStart",
  "agent_id": "a3fecd5",
  "agent_type": "Explore"
}

SubagentStop Hook Input

{
  "session_id": "main-session-uuid",
  "hook_event_name": "SubagentStop",
  "agent_id": "a3fecd5",
  "agent_type": "Explore",
  "agent_transcript_path": "/path/to/subagent/agent-a3fecd5.jsonl",
  "last_assistant_message": "Research complete. Found 3 auth patterns..."
}

AMC Spawn Tracking

AMC tracks spawned agents through:

Pending Spawn Record

// ~/.local/share/amc/pending_spawns/<spawn-id>.json
{
  "spawn_id": "550e8400-e29b-41d4-a716-446655440000",
  "project_path": "/Users/dev/myproject",
  "agent_type": "claude",
  "timestamp": 1708872000.123
}

Session State with Spawn ID

// ~/.local/share/amc/sessions/<session-id>.json
{
  "session_id": "session-uuid",
  "agent": "claude",
  "project": "myproject",
  "spawn_id": "550e8400-e29b-41d4-a716-446655440000",
  "zellij_session": "main",
  "zellij_pane": "3"
}

Resuming Subagents

Subagents can be resumed using their agent ID:

{
  "type": "tool_use",
  "name": "Task",
  "input": {
    "description": "Continue auth research",
    "prompt": "Continue where you left off",
    "subagent_type": "Explore",
    "resume": "a3fecd5"
  }
}

The resumed agent receives full previous context.

Worktree Isolation

For isolated code changes:

{
  "type": "tool_use",
  "name": "Task",
  "input": {
    "prompt": "Refactor auth module",
    "subagent_type": "general-purpose",
    "isolation": "worktree"
  }
}

Creates temporary git worktree at .claude/worktrees/<name>/.

Result includes worktree info:

{
  "type": "tool_result",
  "content": "Refactoring complete.\n\nWorktree: .claude/worktrees/auth-refactor\nBranch: claude/auth-refactor-abc123\n\nChanges made - worktree preserved for review."
}