Skip to content
Claude Code Docs Claude Code Docs

Orchestration & Agent Tools

Agent

Section titled “Agent”
  • Invoked as: Agent
  • Source directory: claude-code-main/tools/AgentTool/
  • Class: Orchestration
  • Side effect: Spawns
  • Gating: Always
  • Documented in public docs?: Yes — https://code.claude.com/docs/en/sub-agents
  • Description: “Spawns a subagent with its own context window to handle a task.” Subagents are specialized helpers that run in isolated context and return results. Use when a side task would flood the main conversation with search results or logs. Define custom subagents in .claude/agents/ or ~/.claude/agents/.
  • Input parameters:
    • agent (string, required) — Subagent name or type (e.g., general-purpose, Explore, Plan)
    • prompt (string, required) — Prompt to send to the subagent
    • tools (string, optional) — Comma-separated tool allowlist for the subagent
    • disallowedTools (string, optional) — Tools to deny
    • model (string, optional) — Model for the subagent (e.g., haiku, sonnet)
  • Returns: Subagent’s response and result.
  • Notes: Subagents have their own context windows and do not inherit the main conversation history (exception: fork subagents — see agents/how-subagents-work#fork-subagents). Built-in agents: general-purpose, Explore, Plan, plus helpers (statusline-setup, Claude Code Guide). Load project or user-defined subagents by name. Subagents cannot spawn other subagents — Agent(...) restrictions in a subagent’s tools field have no effect; only main-thread agents (those running via claude --agent) can spawn subagents. Subagents also cannot use TeamCreate.

Skill

Section titled “Skill”
  • Invoked as: Skill
  • Source directory: claude-code-main/tools/SkillTool/
  • Class: Orchestration
  • Side effect: Spawns
  • Gating: Permission required
  • Documented in public docs?: Yes — https://code.claude.com/docs/en/skills#control-who-invokes-a-skill
  • Description: “Executes a skill within the main conversation.” Skills are reusable prompt-based workflows stored as markdown in .claude/skills/ or ~/.claude/skills/. Use Skill to invoke skills programmatically; users invoke with /skill-name or by mentioning the skill name.
  • Input parameters:
    • skill (string, required) — Skill name (directory name under .claude/skills/)
    • arguments (string, optional) — Arguments to pass to the skill
  • Returns: Skill’s output.
  • Notes: Permission required. Skills are merged inline into the main conversation (unless context: fork is set in frontmatter). Control invocation via frontmatter: user-invocable: false hides from / menu; disable-model-invocation: true prevents automatic invocation.

SendMessage

Section titled “SendMessage”
  • Invoked as: SendMessage
  • Source directory: claude-code-main/tools/SendMessageTool/
  • Class: Orchestration
  • Side effect: None
  • Gating: Feature-flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1; always available when teams are enabled
  • Documented in public docs?: Yes — https://code.claude.com/docs/en/agent-teams
  • Description: “Sends a message to an agent team teammate, or resumes a subagent by its agent ID. Stopped subagents auto-resume in the background. Only available when CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 is set.”
  • Input parameters:
    • recipient (string, required) — Teammate name or agent ID
    • message (string, required) — Message to send
  • Returns: Confirmation that message was sent.
  • Notes: Requires CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 feature flag to be enabled. Used within agent teams for direct inter-agent communication. Teammates auto-resume after stopping if you send them a message.

TeamCreate

Section titled “TeamCreate”
  • Invoked as: TeamCreate
  • Source directory: claude-code-main/tools/TeamCreateTool/
  • Class: Orchestration
  • Side effect: Spawns
  • Gating: Feature-flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
  • Documented in public docs?: Yes — https://code.claude.com/docs/en/agent-teams
  • Description: “Creates an agent team with multiple teammates. Only available when CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 is set.” The lead session spawns independent teammates, each with its own context, and coordinates shared work via a task list and mailbox.
  • Input parameters:
    • team_name (string, required) — Name for the team
    • teammates (object, required) — Map of teammate name to agent type/definition
    • task_description (string, required) — Description of the work to coordinate
  • Returns: Team ID, teammate IDs, and shared task list location.
  • Notes: Experimental and disabled by default. Requires CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. Teams are stored locally at ~/.claude/teams/{team-name}/. Each teammate is an independent Claude Code session with its own context window. Teammates can message each other and claim tasks from the shared list.

TeamDelete

Section titled “TeamDelete”
  • Invoked as: TeamDelete
  • Source directory: claude-code-main/tools/TeamDeleteTool/
  • Class: Orchestration
  • Side effect: Executes
  • Gating: Feature-flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
  • Documented in public docs?: Yes — https://code.claude.com/docs/en/agent-teams
  • Description: “Disbands an agent team and cleans up teammate processes. Only available when CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 is set.” Must be called from the team lead; teammates cannot self-cleanup.
  • Input parameters:
    • team_name (string, required) — Name of the team to delete
    • force (boolean, optional) — Force cleanup even if teammates are still running
  • Returns: Confirmation that team was deleted.
  • Notes: Experimental. Must be called by the team lead, not a teammate. Checks for active teammates and fails if any are running unless force: true. Cleans up team config and task list files.

← Back to Tools/README.md