Chat With Agent

Chat with Agent

When to Use

Use this skill when you need to ask another agent a question, seek help, request a plan, request a review, request decision support, or engage in any form of communication. If the user explicitly asks to talk to a specific agent, you should also use this skill.

Should Use

• You need another agent's expertise, judgment, or second opinion
• You need to request a plan, review, or recommendation from an agent
• The user explicitly asks a specific agent to participate, assist, or answer
• You need to continue an existing agent session while preserving context

Should Not Use

• You can complete the task yourself and the user has not explicitly asked to involve another agent
• It is a simple Q&A that does not require a specialized agent
• The target agent is unclear — you should ask for clarification or list available agents first
• You just received a message from an agent and are about to immediately call back the same agent, which may cause a loop

Decision Rules

1. If the user explicitly requests a specific agent, follow that request — but still look up the agent first; do not guess the ID
2. If you can complete the task yourself, do not call another agent
3. When continuing a conversation, you must pass session_id
4. By default, prefer using list_agents() and chat_with_agent(...) for foreground conversations — do not resort to other methods
5. If the task should run in the background, use submit_to_agent(...) to submit it, then check_agent_task(...) to check the status

Usage Flow

Follow this flow strictly when using this skill:

1. Ensure your tool list includes both list_agents() and chat_with_agent(...) built-in tools

• These two tools are the foundation for communicating with other agents — do not remove or disable them
• If you do not have these tools, inform the user that you need them to talk to other agents, and ask the user to add them

2. Use list_agents() to view the currently available agents, and select one by extracting its ID

• Choose the most appropriate agent based on the user's needs and each agent's description
• If no suitable agent is found and you are not the Default Agent, use the Default Agent
• Otherwise, inform the user that no suitable agent is available, and suggest they create a new agent or adjust existing agent descriptions for better matching

3. Call chat_with_agent(...) to initiate a foreground conversation. Key parameters include:

• to_agent: the target agent's ID (note: this is the ID, not the name)
• text: what you want to say to the target agent
• session_id: (optional) if you need multiple rounds of conversation with the same agent, pass the same session_id starting from the second round to maintain context continuity
• timeout: (optional) estimated time needed for the foreground wait, to avoid premature timeouts

4. If the task is better suited for background execution, use the background tool path:

• submit_to_agent(...): submit a background task — requires to_agent, text, and optionally session_id / task_timeout
  • Default execution budget is 3600 seconds (1 hour) when task_timeout is omitted
  • For work that may exceed 1 hour, pass an explicit positive task_timeout in seconds
  • Numeric strings such as "7200" are accepted
• check_agent_task(...): check task status by task_id; returns the final result when complete
  • On timeout the task finishes as failed with an error like Task timed out after Ns (and code=timeout in the API payload)

Minimal Call Examples

New Conversation

list_agents()

chat_with_agent(
  to_agent="<target_agent_id>",
  text="[Agent <your_agent_id> requesting] I need your help determining the approach for this issue.",
)

Continuing an Existing Conversation

chat_with_agent(
  to_agent="<target_agent_id>",
  text="[Agent <your_agent_id> requesting] Please continue expanding on point 2 based on the previous conclusion.",
  session_id="<previous_session_id>",
)

Background Submission and Status Check

submit_to_agent(
  to_agent="<target_agent_id>",
  text="[Agent <your_agent_id> requesting] Please complete this longer task in the background.",
  task_timeout=7200,
)

check_agent_task(
  task_id="<task_id>",
)

Important Notes

• Add a conversation identifier: it is recommended to start your message with:

[Agent <your_agent_id> requesting]

This helps the other agent clearly identify who is speaking, improving communication efficiency and accuracy.

• Reuse sessions wisely: if you need multiple rounds of conversation with the same agent, be sure to pass the same session_id to maintain context continuity. Otherwise, each call is treated as a new conversation, and the other agent may not correctly understand your needs. You can obtain the session_id from the first round's response, for example:

  [SESSION: xxx]
  

  where xxx is the session_id value. This value is typically system-generated and has a long, unique format. Save this session_id and continue using it in subsequent conversations with the same agent.