Documentation Index
Fetch the complete documentation index at: https://docs.oktolabs.ai/llms.txt
Use this file to discover all available pages before exploring further.
Multi-agent development
You can run a multi-agent board by giving each agent a narrow role and letting Pulse enforce the boundaries. A simple production setup is Spec + Executor + Validator.Team structure
| Agent | Preset | Owns |
|---|---|---|
| Spec agent | Spec | Ideation, refinement, spec content, sprint planning, card breakdown. |
| Executor agent | Executor | Card implementation through validation. |
| Validator agent | Validator | Spec, sprint, and task gates. |
Setup
Create the agents in the web UI, assign presets, then export MCP config.Workflow
- Spec agent creates ideation and asks ambiguity-killer questions.
- Spec agent derives or writes the spec.
- Validator submits spec validation when evidence is ready.
- Spec agent derives cards and a sprint.
- Executor implements cards and moves them to
validationwith conclusions. - Validator submits task validation and routes each card.
Preventing cross-role drift
The important rule is that implementers do not validate themselves. Executor can read validation records but cannot submit them or movevalidation -> done.
Board-level overrides
Use board overrides to restrict an agent on a sensitive board. For example, keep a shared Reporter read-only on production boards while allowing broader reads on evaluation boards.Common failure modes
| Symptom | Cause | Fix |
|---|---|---|
| Executor gets permission denied on task validation | Executor lacks card.validation.submit. | Ask Validator to review. |
| Spec agent cannot move approved spec to validated | Gate transition is Validator-owned. | Submit spec validation as Validator. |
| Agent cannot act in current status | Missing interact_in flag. | Adjust preset or board override. |