Brief yourself from Claude
You will need: Alethman connected to Claude Code, Claude Desktop, Cursor or another MCP client (Connect a client), on a key with intel:read and reports:read.
Once connected, you ask in plain words and the assistant picks the tools. Knowing which tool answers which question helps you ask well and check what it did. Every client shows the tool calls it makes; open them to see exactly what it read.
A morning brief
Section titled “A morning brief”What’s moving in my workspace since yesterday? Start with the sectors, then the three stories carried most widely, and tell me where outlets disagree.
The assistant usually calls:
get_workspace, to learn your sectors and states.list_outlooks, for each sector’s level, direction and verdict.list_storieswithhours: 24andorder: "voices".get_storyfor the top three, whosesummaryandbloc_findingsay where coverage agrees and where it divides.
One sector, in depth
Section titled “One sector, in depth”Why is the energy outlook where it is? Which stories drive it, and what forecasts are open?
list_outlooks gives the verdict and its drivers (each with a story_id); get_story opens each driver; list_forecasts with sector: "energy_power" and status: "open" lists the dated forecasts with their probabilities and deadlines.
A state
Section titled “A state”Summarise the last three days in Maharashtra, highest risk first. Leave out anything carried by fewer than five outlets.
list_stories with state: "maharashtra", hours: 72, order: "risk"; the assistant filters on independent_outlets itself.
Reading a report
Section titled “Reading a report”Read me the latest weekly report’s judgements and anything it says about public order.
list_reports, then get_report with the newest id. Reports come as Markdown a page at a time; the assistant asks for page: 2 and on as it needs them. Each page ends with “Page n of m”.
Coordination
Section titled “Coordination”Are there any grade A coordination findings among the latest 60? Who does the reporting name as orchestrating them, and through which outlets?
list_coordination with grade: "A" and limit: 60. The limit is applied before the grade filter, so with the default of 10 the tool looks at the latest ten findings only; asking for 60 makes it look further back. Each finding names its orchestrator, the mediums carrying it hardest and a target, from the roles the reporting gives each party, with the stories it rests on. A grade says how much the reporting supports a pattern, not that anyone coordinated it.
The articles themselves
Section titled “The articles themselves”Find this week’s articles mentioning “Tarali” and list the outlets that ran them.
list_signals with q: "Tarali" and since a week back. It returns 25 articles at a time; the assistant pages on with next_cursor if you ask for more.
Getting good answers
Section titled “Getting good answers”- Ask for sources. “Cite the story ids” or “link each claim to a story” makes the assistant say which
get_storyresult each sentence came from. Story ids open in the dashboard too. - Name the window. “Since yesterday”, “this week”, “the last 72 hours” become
hoursorsince. Without one,list_storiesreads 72 hours. - Name the order. “Most widely carried” is
order: "voices"; “most serious” isorder: "risk"; “latest” isorder: "recent". - Keep lists short. A tool result is capped at 40,000 characters. If a result says
truncated, ask for fewer, or narrow by sector or state. - It reads only your workspace. Stories outside your sectors and states are invisible to it, as they are to the API.
Saving a prompt
Section titled “Saving a prompt”In Claude Code, save a brief you run every morning as a slash command: put the prompt in .claude/commands/brief.md and run /brief. In Claude Desktop, keep it in a Project’s instructions.