Explicit termination flags, sometimes called stop sequences, exit conditions, or termination signals, are control directives defined within a Zia Agent's workflow that ensure execution halts deterministically once a task is complete, a boundary is hit, or human intervention is genuinely needed. These conditions exist to prevent loops, avoid unnecessary token expenditure, and enforce safety guardrails consistently.

1. Task Completion and Yield Flags

When building custom agents in Agent Studio, developers use explicit terminal steps to signal that the agent has genuinely fulfilled its objective.

• Explicit output signals. Declarative end-status flags, such as COMPLETE or SUCCESS, passed to the execution runtime to indicate the goal has been achieved, typically by returning a final structured response without invoking any further tool calls.
• Handoff termination. When an agent passes control to another agent within a multi-agent network, or hands off to a human representative, an explicit handoff flag breaks the current agent's active execution loop, ensuring it stops rather than continuing to run alongside whoever it just handed off to.

2. Guardrails and Safety Policy Violations

Agents enforce hard guardrail boundaries set during their initial configuration.

• Scope and tool permission flags. If an agent attempts an unauthorised action, such as calling an unapproved API, accessing restricted CRM or Desk records, or attempting an administrative deletion, a policy violation flag is raised that immediately terminates the session.
• Content filter and compliance violations. Explicit flags terminate the agent's process if inputs or outputs breach internal safety guidelines or Zoho Zia's Terms of Use, such as a security policy violation or an attempt to execute an untrusted payload.

3. Stop Sequences and LLM Token Termination

At the underlying model parameter level, whether using Zoho's own Zia LLM or a connected BYOK or third-party model in Agent Studio, two further mechanisms apply.

• Explicit stop sequences. Custom strings or token markers, such as [END], </s>, or "Observation: Stop", defined in the model's settings, tell the underlying model to stop generating text at that point.
• Max turn or iteration limit. A safety circuit breaker that triggers once an agent reaches a predefined number of action or reasoning turns without reaching a resolution, preventing an infinite tool execution loop from running indefinitely.

4. Platform-Level and Account Termination Controls

Beyond individual agent logic, two broader controls exist at the platform level.

• Manual deactivation flag. Disabling an agent moves it directly into the Draft state and invalidates any of its currently active execution triggers.
• Portal disabling. Disabling the Zia Agent Portal entirely emits a platform-wide termination signal that forcibly halts every active and scheduled agent job immediately, across the whole organisation rather than a single agent.

How to Implement Explicit Termination in Your Own Agent Workflows

1. In Agent Studio's node or workflow view. Define clear end-state nodes on every logical branch. Make sure conditional branches, such as "If Action Fails" or "If Information Found", terminate gracefully into an output response node, rather than routing back into an operational tool and risking a loop.
2. In natural language prompts. Include explicit termination instructions directly in the agent's system prompt, for example: "If the required ID is missing from the record, reply with '[TERMINATE: MISSING_DATA]' and do not call any further tools."
3. In multi-agent setups. Ensure every sub-agent returns an explicit completion status to the parent orchestrator agent before closing its own context window, so the orchestrator always knows definitively whether a sub-agent finished successfully or not.