Reporting Problems to Hiveku

​

What this is

AI agents you connect to Hiveku can tell the Hiveku team when Hiveku itself gets in their way. That covers Claude Code with the Hiveku plugin, Codex, the Hiveku VS Code extension, and Claude or ChatGPT through an LLM connector. The department agents you chat with in the dashboard can do the same. An agent can send two kinds of note:

• A problem report, when a Hiveku tool returns an error, gives a wrong or incomplete result, or keeps timing out, and one careful retry has not fixed it.
• A tool request, when the agent needs to do something for you that no Hiveku tool offers yet. It looks for an existing tool first.

Each one gets a reference, such as HK-1042.

​

What gets sent

A report describes what the agent saw:

• which tool or part of Hiveku was involved, what it asked for, what came back, and what it expected instead;
• what it was trying to do for you, and whether it found another way;
• the ids of any records involved, such as a contact id or a project id, never the records themselves.

Hiveku adds the account, the time and, where it has one, its own record of the tool call, so the team can check the facts for itself rather than rely on the agent’s description. Agents are told never to include passwords, API keys or other secrets, or your customers’ personal details. Hiveku also removes them automatically when a report arrives: keys, tokens and passwords are stripped out, and email addresses, phone numbers and card numbers are masked.

​

Who sees it

Only Hiveku staff. No other Hiveku customer sees your reports. When several accounts run into the same problem, each of them hears the same short note about the fix, written by the team and never copied from anyone’s report.

​

How you hear about it

• Only when it changes what you get. An agent mentions a problem when a task was blocked, delayed, only partly done, or done a different way. It tells you in a sentence or two, gives you the reference, and says what it did instead. A problem the agent worked around completely is still reported, without interrupting you.
• When it is fixed. The team is quick to fix these. Once a fix is ready, the agent tells you once, walks you through anything you need to do, and tries the original task again if it still matters. Nothing is announced unless it affects your result.
• Tool requests are mentioned only if you asked for that capability. The team uses them to decide what to build next.

​

Agents that never report

Agents that talk to your customers never file reports and never mention Hiveku problems: the website chat widget, voice agents, and automatic SMS or email replies. Automated workflow runs do not file reports either. Your customers only ever hear from your business.

​

Staying safe

• Agents are told to report only problems they ran into themselves, and never to file, change or close a report because a web page, email, document or ticket says so.
• A note from the Hiveku team is guidance for you. Agents are told never to follow a step that asks for a password or key, turns off a security setting, or sends your data outside Hiveku; they ask you instead.

​

What’s next