Implementation workbook / Edition 2.0 / Reviewed 16 September 2026
Use the fictional examples to practice. Replace placeholders only with approved facts, then complete the checks before using a workflow with real customers.
Inside this workbook
Use Kimi or another assistant without surrendering control
The worksheets in this edition use ordinary text and explicit inputs, so you can start in an assistant your organization already approves. A chat draft is enough for one email or one checklist. Use an agent only when the task actually needs files or multiple steps. Tool subscriptions, credits, and configuration are separate from this CreatorHQ kit.
Kimi example: its current help center describes Agent tools for documents, research, websites and other file-oriented work, and lists K3 in its model guidance. Availability and limits depend on the actual account and mode. Check Kimi's Agent overview and model-selection help, rather than assuming an older tutorial's model selector still applies.
Bounded task: “Using only the fictional sample data and approved template I provide, create one editable worksheet for [task]. Include a separate assumptions-and-checks section. Do not browse private accounts, send messages, publish, make purchases, connect integrations, or change source records. Ask before any external action. Stop when the file and validation checklist are ready.”
Cost control: check the account's available credits and expected task scope before starting. Kimi documents a shared credit pool with usage-based consumption. Do not launch repeated large tasks or a swarm for a simple rewrite. Read the current billing rules; this guide does not promise a free tier or a fixed cost per result.
Acceptance test for any provider: use the same fictional input twice; inspect whether facts, totals, links and limits survive. Open the returned file in the intended app. Try a missing required fact and a conflicting instruction embedded in source text. The right result is a flagged uncertainty or request for approval, not an invented fact or unauthorized action. Record the provider, mode, date, result, and actual usage so your choice reflects your workflow rather than a leaderboard.
Human review remains necessary: compare output to the source, check permissions, and approve every customer-facing or money-related action. Agent capabilities and limits change; recheck the official feature and limit guidance when you revise this workflow. Sources reviewed 16 September 2026. Kimi is an example, not an endorsement of its suitability for private or regulated data.