Skip to main content

Explyt configuration reference

Pick a mechanism based on your task. The individual pages describe file paths, scope, limitations and ways to verify the setup. Where a user-facing contract has not been published yet, the documentation says so explicitly.

Which mechanism to choose

TaskMechanismWhere to read
Give the agent general information and project rulesAGENTS.mdAGENTS.md
Apply an instruction to specific filesRulesRules
Capture a repeatable working scenarioSkillsSkills and the SKILL.md schema
Keep durable knowledge across chatsMemory BankMemory Bank
Assign different models to subagentsModel RoutingSubagent profiles
Connect an external service or toolMCPMCP
Restrict reading, writing and running actionsEdit Scope, ignore files and permissionsPermissions and safety

Text instructions in AGENTS.md, Rules, Skills and Memory Bank do not replace technical access restrictions. If an instruction asks for an unavailable action, the agent mode, Edit Scope, .readignore, .writeignore and tool permissions take precedence as technical restrictions.

Interface and current chat settings

If a setting does not apply

  1. Check the selected mode, Edit Scope, the ignore files, the model and the MCP status.
  2. Keep the instruction you are testing in a single source: the current request, a Skill, a Rule or AGENTS.md.
  3. Start a new chat to rule out the influence of old history.
  4. Bring the sources back one at a time and note which step changes the behavior.

Detailed checks for each mechanism are collected on the Configuration troubleshooting page.