Promethist for Claude

Operating Guides

Claude does not come with knowledge of Promethist built in. Before it works in an area, it loads the matching operating guide from the server with the get_guide tool. The guides describe how the platform behaves, which fields mean what, what makes a good agent or evaluation, and the mistakes to avoid. They are written and updated by Promethist together with the platform, so Claude always works from the current state without anything to update on your side.

You can ask for a guide explicitly — "read the evaluation guide first" — when you want Claude to be thorough about a topic.

Topic What it covers

Topic	What it covers
`domain`	How Promethist fits together: accounts, projects, agents and their revisions, packages, project assets.
`workspace`	Accounts and projects, the Owner, Editor and Viewer roles, members and invitations.
`authoring`	Writing agents: their fields and limits, completeness rules, voices and avatars, templates, guardrails and business processes.
`quality`	The rubric Claude uses to judge whether an agent or a project is ready.
`packages`	Project packages, and how to build an Empower coaching project and an Engage kiosk.
`knowledge`	Knowledge sources, adding and refreshing web pages, how retrieval behaves, referencing knowledge from agents.
`multimodal`	Interactive content: its types, handovers between agents, and referencing content from agents.
`integration`	Webhooks before and after a conversation, MCP servers and connectors, and choosing which tools an agent gets.
`skills`	What a skill is, how an agent uses it, and why one may stay unused.
`evaluation`	Designing evaluations: output data types, groups, the featured metric, when evaluations run, where results go, and how revisions are published.
`analytics`	What each analytics figure means, its unit, and how it is commonly misread.
`data-export`	Where transcripts, raw turns and model logs come from, and what each source cannot tell you.
`billing`	Conversational minutes, account usage, plans and invoices.
`testing`	Planning and running live test conversations with an agent and turning failures into fixes.
`test-automation`	Building simulated users, running them as test suites and reading the results.
`memory-testing`	Running several sessions as one returning user, spreading them over time, and re-running the session under test from a memory snapshot.
`relational-metrics`	The evaluations a tester is judged by, how to read their pass rates, and why the older score-based sets mislead.

domain

How Promethist fits together: accounts, projects, agents and their revisions, packages, project assets.

workspace

Accounts and projects, the Owner, Editor and Viewer roles, members and invitations.

authoring

Writing agents: their fields and limits, completeness rules, voices and avatars, templates, guardrails and business processes.

quality

The rubric Claude uses to judge whether an agent or a project is ready.

packages

Project packages, and how to build an Empower coaching project and an Engage kiosk.

knowledge

Knowledge sources, adding and refreshing web pages, how retrieval behaves, referencing knowledge from agents.

multimodal

Interactive content: its types, handovers between agents, and referencing content from agents.

integration

Webhooks before and after a conversation, MCP servers and connectors, and choosing which tools an agent gets.

skills

What a skill is, how an agent uses it, and why one may stay unused.

evaluation

Designing evaluations: output data types, groups, the featured metric, when evaluations run, where results go, and how revisions are published.

analytics

What each analytics figure means, its unit, and how it is commonly misread.

data-export

Where transcripts, raw turns and model logs come from, and what each source cannot tell you.

billing

Conversational minutes, account usage, plans and invoices.

testing

Planning and running live test conversations with an agent and turning failures into fixes.

test-automation

Building simulated users, running them as test suites and reading the results.

memory-testing

Running several sessions as one returning user, spreading them over time, and re-running the session under test from a memory snapshot.

relational-metrics

The evaluations a tester is judged by, how to read their pass rates, and why the older score-based sets mislead.

The list of topics is part of the get_guide tool’s description, so Claude always sees the topics the server currently offers. A topic marked "(Promethist staff only)" is internal material that the server serves only to Promethist staff.