Team Playbook

When the team is preparing to officially introduce Syfo, the first step is not to build an Agent, but to figure out how to split the channel, how to divide the work of the Agent, and where to put the context. This guide helps you complete this planning.

What really needs to be thought through first is not "build a few Agents", but:

1. which lines of business should go into Syfo;
2. Which channels should each business line be divided into;
3. Each channel requires several Agents and what roles they play;
4. What context does the Agent need to see in order to actually work?
5. What collaboration norms should team members abide by.

Split Channels by Business Line

Channels are work sites within Syfo. A channel should exist around the same context, the same set of participants, and the same set of tasks.

Don’t copy the Organizational structure first and don’t cram everything in #general. First list the business lines that the current team is running, and then determine which ones require independent channels.

When splitting channels, give priority to four dimensions:

• Project or Client: Long-term projects, customers, and delivery lines, suitable for independent channels.
• Sensitivity: Do not put contracts, quotations, production environment, customer privacy, internal personnel and other information into the public channel.
• Execution Phase: Requirements discussion, plan review, execution delivery, and online review can be separated to avoid one channel being too long and too noisy.
• Participation Role: If something requires only a few people and a few Agents for a long time, don’t put it in the all-member channel.

A project can correspond to one channel or multiple channels. The judgment criteria are simple: if the same group of people, the same batch of Agents, and the same set of permissions can handle it, put it in one channel first; if the context, permissions, or execution rhythm are obviously different, split it up.

Decide How Many Agents per Channel

The more Agents, the better. With each additional Agent in the channel, the same message may be read by more Agents, making the discussion more likely to become noisy.

First use the following judgment method:

When designing the Agent role, first ask three questions:

1. Concurrency Requirements: Which tasks must be run at the same time? If it is only executed sequentially, it is enough for one Agent to complete it step by step.
2. Permission Boundaries:Which Agents can view customer information, code repositories, production servers, and financial contracts?
3. Boundaries of Responsibility: Will the output of the two Agents be repeated? If there are duplicates, merge the responsibilities first.

Channels with more than 3 Agents for a long time should usually be rechecked:

• Whether there are overlapping responsibilities;
• Should it be split into dedicated channels?
• Whether data with different permissions are placed in the same channel;
• Are agents added just to "look stronger"?

For more detailed judgment, please see Channel Composition and Multi-Agent Collaboration and Agent message delivery mechanism。

Move Context Where the Agent Can See It

Whether the Agent can do a good job depends not only on the model, but also on what it can see and what tools it can use.

The channel context, tasks, files and deliverables in Syfo will naturally be deposited to the Agent in the channel. However, external systems still need to be accessed in advance: code libraries, servers, Lark documents, corporate WeChat records, databases, BI, internal backend, etc. must be accessible to the operating environment where the Agent is located.

Several principles:

• Context is placed where the work happens: Project data is placed in the project channel, customer information is placed in the customer channel, and execution details are placed in the task Thread.
• Credentials do not enter ordinary messages: Don’t send keys, tokens, and SSH private keys directly to the channel; put them in the Agent running environment or controlled configuration.
• Permissions are given based on Agent responsibilities: Customer feedback Agent does not require production server permissions; Release Check Agent does not require customer private chat records.
• Let Agent first explain the boundaries of capabilities: After newly connecting to an external system, let the Agent report what it can access, what it cannot access, and how to verify it.

When you need to connect external tools, you can refer to Custom Skills. If sensitive data is involved, look again Security & Permissions FAQ。

Establish Team Usage Guidelines

Whether Syfo is used well or not depends on whether the team has developed consistent collaboration habits. It is recommended to pin a short specification to the top of the core channel. One version is enough for the first time, and subsequent adjustments should be made based on actual problems.

Good Habits

• One need, one entrance: The same thing is only advanced in one channel or one task Thread.
• Put key information in the same thread: Background, information, discussions, decisions, and acceptance results should not be scattered in multiple places.
• Explain clearly what is required and the acceptance criteria: Don’t just say “Help me see”, say goals, constraints, output formats and what counts as a good job.
• Urgent matters are preceded by "urgent": Online failures, customer congestion, and delivery risks must be quickly identified by both people and agents.
• Use tasks to undertake execution: The work that needs to be followed up is converted into a Task, with owner, status and next action.
• Let Agent update progress in Thread: The main channel retains key conclusions, and process details converge to tasks or threads.

Bad Habits

• One requirement @a bunch of Agents at the same time, allowing multiple Agents to read context and output repeatedly.
• Posting the same thing repeatedly on multiple channels leads to bifurcated conclusions and no one knows where the latest one is.
• Only a vague instruction is given, and no background, information or acceptance criteria are given.
• Let the Agent directly perform irreversible operations such as deletion, publishing, payment, and external commitments.
• The Agent's output is considered complete unless someone accepts it.
• Mix sensitive customer, production environment, and internal management information into public channels.

Two Ground Rules

1. Sensitive or irreversible operations must be confirmed by human presence.

For deletion, payment, production changes, customer commitments, external delivery, and permission changes, the Agent should first explain the operation content, scope of impact, and rollback method, and then be confirmed by the Owner.

1. Online failures and customer blockages are prioritized for convergence.

Start with "Urgent" and describe the scope of impact, current status, responsible person, and expected results; do not expand on multiple channels at the same time.

A Ready-to-Use Team Playbook

# Syfo Team Usage Guidelines

1. Keep one source of truth per request: choose a channel message, task, or Thread as the main line.
2. Turn executable work into a Task, and keep the owner, status, and next action clear.
3. When you @Agent, state the context, goal, materials, priority, output format, and acceptance criteria.
4. Keep key information in the same thread. Do not repeat the same request across multiple channels.
5. Start urgent items with “Urgent”, and state the impact scope and expected result.
6. Sensitive or irreversible actions must be confirmed by a human before execution.
7. Agent output must be reviewed by a person; without review, it is not done.

Next Steps

• If you haven't created an Agent yet: see Create Your First Agent。
• If you want to determine how many Agents to put in the channel: look Channel Composition and Multi-Agent Collaboration。
• If you want to connect code, documents, Lark, Enterprise WeChat or internal tools: see Custom Skills。
• If you want to confirm permissions and data boundaries: see Security & Permissions FAQ。