Skip to main content
Click New Agent to open the two-step creation flow. It is intentionally short: create the teammate first, then tune the machinery from the agent page. Before you start, skim Concepts - What Goes Where. It explains the four layers of an agent’s brain: system prompt, skills, memory, and wiki.

​
Step 1: Identity

Identity defines how the agent appears in the dashboard and how it fits into your AI team.
FieldWhat it is
Agent nameDisplay name shown in Slack and the dashboard
RoleSpecialist does the work; Boss delegates to specialists
Reports toFor specialists, choose one or more Boss agents
TagsOptional labels for dashboard filtering and team organization
Boss agents skip persona selection. Their operating prompt is generated from the team registry so it stays synchronized with the specialists that report to them.

​
Step 2: Profile or Launch

For specialist agents, choose a starting brain:
OptionUse when
PersonaYou want a strong starter prompt and skills for a common role
BlankYou know the job and want Coach to draft the details
Import configYou have an exported agent config with prompt and skills
For Boss agents, Step 2 is a launch review. Create the boss, then add specialists and let the team registry drive the orchestrator prompt.
The creation flow does not need to capture every rule. After creation, open Coach and describe what you want in plain English. Coach can draft or refine system prompts, skills, memory, evaluation cases, and operating rules.

​
Configure after creation

The creation flow sets the agent identity and starting profile. Everything operational is configured from the agent page after creation.
AreaWhere to configure it
Slack connectionAgent page → Overview
ModelAgent page → Overview
MCP serversAgent page → Tools
Internet / shell accessAgent page → Tools
System prompt, skills, memoryAgent page → Instructions
Wiki foldersAgent page → Wiki
Access controlAgent page → Settings / access controls
Sensitive-data monitoringAgent page → Settings → General
Evals and regression casesAgent page → Settings → Evals
Test conversationAgent page → Test
This keeps creation fast while keeping each setup area easy to find.

​
After creation

SlackHive opens the agent page. A typical setup sequence is:
  1. Connect Slack - create or reuse a Slack app, then save bot credentials.
  2. Tune with Coach - ask Coach to build or improve the system prompt and skills.
  3. Assign tools - add MCP servers and capabilities from the Tools tab.
  4. Assign knowledge - attach shared Knowledge Library folders.
  5. Set safety controls - configure sensitive-data monitoring and redaction for agents that touch private data.
  6. Test and evaluate - use Test Mode, then add eval cases for behavior that must not regress.
  7. Invite to Slack - invite the bot and start mentioning it.

​
Agent page tabs

TabWhat you manage
OverviewIdentity, model, Slack connection, status, satisfaction, configuration
InstructionsSystem prompt, skills, and memory
ToolsMCP servers, internet access, shell access
WikiShared Knowledge Library folders assigned to this agent
AudiencesGroup-specific prompt overrides
SettingsAccess and operational settings
Changes hot-reload within seconds. You do not need to restart SlackHive after editing an agent.

​
Specialist vs Boss

Agent typeBehavior
SpecialistReceives direct mentions, uses tools, answers users, can report to one or more Boss agents
BossReceives high-level requests, delegates to specialists in the Slack thread, and summarizes the outcome
See Boss Agents for delegation patterns and team registry behavior.

​
Next steps

Coach

Build prompts, skills, memories, and operating rules through chat.

AI Backends

Choose whether agents run on Codex/OpenAI or Claude/Anthropic.

Knowledge Library

Build shared wiki folders from repos, files, or URLs.

Tools & Permissions

Configure MCP servers, internet access, shell access, and tool safety.

Agent Evals

Add healthchecks and regression cases before going live.

Sensitive Data

Configure detection, redaction, and audit workflows.