The agent page

The page you see when you open one of your runs. Its title is the run's name, with the kind of agent under it in italics; the pencil beside it renames the run (display only). Its parts, from top to bottom:

Console

A conversation, not a log:

• What the agent says appears as short messages. A long one is folded with Show more; the closing message of a finished run is its Summary, with a link to the full report.
• What the agent does in between is folded into one line, "Ran 6 steps". Open it to see each step, and open a step to see what it returned. Failed steps are counted in red. While a run is live the line reads "Working · …".

• What you or a teammate said to the agent appears as a message too.

• Send puts your message in the agent's queue. It arrives after the agent's next step. A message never interrupts a step that is running, so a scan is never left half-done.

• Note is saved in the run's record and its report. It never reaches the agent.
• Attach gives the run a file. A run that is going gets it after its current step; a finished run's file goes with your next run. See Attaching files.
• Neither one can widen what the agent may do. The tool allow-list and the scope are enforced outside the conversation, where nothing you type can change them.
• The agent's narration is its own account of what it's doing. It is useful, but it is not a trace of how the agent actually reached its answer.

The buttons beside the message box:

If the agent asks you a question, a yellow banner shows it. Answer with Send. The agent waits for your answer for up to 24 hours, and the time it spends waiting does not use up the run's time limit (that limit counts only the time the agent is working). If nobody answers within 24 hours the run ends as incomplete, and its report names the question that was left unanswered. If the agent asks to use an extra program, Allow for this run applies to this run only. Scope and safety checks still apply to every command.

Report and marking a run complete

Report shows this run's report, and lets you open it full-page, save it as PDF or download it as a Word file. When the run has finished, Mark complete moves it to Task History on the dashboard, which lists every run you have completed. This page is about one run only; there is no list of earlier runs here.

Tools & permissions

Each tool and program is listed by name, with a short description of what it does.

• Scope boundary (for example CIDRScope) is the kind of target the agent works on. A target you give it can only narrow the engagement's authorized scope, never widen it.
• Tools are the actions the agent itself can take.
• Programs it may run is the agent's command-line allow-list. It can run these programs and no others, and every command is checked against the scope first.
• Always refused lists patterns that are blocked even inside an allowed program, such as shell chaining or redirection.

The agent never gets a raw shell.

Engagement

Your organisation's authorization record: the client, who authorized the work, the dates, and the scope this agent may use. Every agent shares it. Admins edit and approve it under Settings → Engagement.

My preferences

Notes the agent reads at the start of each run you launch, for example how you like reports laid out. They affect style only, never scope, tools or limits. Each user has their own preferences, and an admin can view yours but not change them.

The agent may suggest something to remember. Nothing is saved unless you choose Add to my preferences.

Team chat

See Teams. On a team page, Send reaches every member after its next step. A team member's page has a Team link back to the group chat.