Agents and permissions
You can use agents to divide ownership across spec writing, implementation, validation, QA, and reporting. Permissions make that division enforceable at the tool layer.
What is an agent in Pulse
An agent is an API-key-backed identity with a name, objective, description, active flag, board access, and resolved permission set. The MCP server authenticates the key and checks permissions before executing tools.
The permission model
Pulse uses dotted granular flags such as spec.move.draft_to_review, card.validation.submit, and kg.query.decision_history. The current registry contains 270 enabled leaf flags in source.
Five built-in presets
The core operating roles are:
| Preset | Use it for |
|---|
| Full Control | Solo or trusted admin work. |
| Spec | Ideation, refinement, spec content, sprint planning, and card breakdown. |
| Executor | Normal card implementation through validation, without gate submission. |
| QA | Test scenarios and test card lifecycle. |
| Validator | Spec, sprint, and task gates. |
Board roles vs agent-level flags
Agent-level flags define the maximum capability. Board-level overrides restrict that capability on a specific board; they do not expand it. This ceiling model prevents a board override from turning a narrow agent into a broader one.
Managing agents
Use the web UI for day-to-day setup. Use the CLI to export MCP config after agents exist:
Wrote .mcp.json with MCP server entries for available agents.
Related pages
Last modified on May 8, 2026
