Myskillos — AI skill & agent platform

← Red vs Blue: Multi-Agent Security Design ReviewGuest mode — changes not saved

Files

Drag to canvas

Agents

👑Chief

🧩Sub-agent

Workflow patterns

Orchestration nodes that govern child agents

📝Sequential

⚡Parallel

🔁Loop

How do agents communicate?

Sub-agents do NOT talk to each other directly; each runs in an isolated context. Coordination happens via (1) the chief — it collects results and feeds one agent's output to another; (2) shared files — agents read/write under workspace/ (a blackboard). Parallel agents run concurrently but isolated; communication happens at the synthesis (merge) point.

Connection types (click an edge to change)

delegate — chief calls the agent via the Agent tool; result returns automatically
parallel — chief → launches concurrently, isolated
handoff — passes output to the next
report-to — returns result to the chief
shared — coordinates via a shared file

A Claude skill is a folder: SKILL.md (name+description frontmatter + instructions). Specialist work is delegated to sub-agents in .claude/agents/*.md — each with its own context window and tools. The best target for true multi-agent skills.

👑Identity

Chief

Name

lowercase-dash format (REQUIRED)

Role

Descriptionsubagent trigger

Folder / category

.claude/agents/<category>/<agent>.md — leave empty for the root folder.

Scope

Model & reasoning

Model

Effort

Estimated cost$$

Rough estimate: model × effort. $ economical · $$ balanced · $$$ expensive.

haiku → fast scan · sonnet → analysis · opus → complex planning · inherit → same as parent

Tool access

Allowed tools

Leave empty to inherit all tools.

Disallowed tools

Permission & memory

Permission mode

Memory

System prompt

…