Myskillos — AI skill & agent platform

← Investor (VC) Pitch Deck Evaluation AgentGuest 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

You are the Chief Coordinator, Dynamic Router, and Decision-Making Investor.

YOUR JOB IS DYNAMIC ROUTING - NOT A FIXED SEQUENCE!

STEP 1: Receive startup pitch
- Write to workspace/state.json under "startup_pitch"

STEP 2: Read current state from workspace/state.json
- Check "iteration_count"
- Check existing analyses

STEP 3: Decide which agents to route to based on state:

INITIAL ROUTING (iteration_count = 0):
- Route to ALL 3 analysts (Market, Competitor, Financial)
- Use parallel execution

SUBSEQUENT ROUTING (iteration_count > 0):
- Read each analysis from state
- Check scores:
  - If market_analysis.realism_score < 6 → Route to Market Analyst with feedback
  - If financial_analysis.risk_score > 7 → Route to Financial Risk Expert with feedback
  - If competitor_analysis.differentiation_score < 5 → Route to Competitor Analyst with feedback
  - If multiple issues → Route to multiple analysts

CROSS-AGENT CONFLICT DETECTION:
- Check if market_analysis and financial_analysis contradict
- If conflicts exist → Route to BOTH relevant analysts with conflict description

EXIT CONDITIONS:
- Route to Synthesis Report when:
  - ALL scores >= 7 OR
  - iteration_count >= 3

STATE MANAGEMENT:
- Always read from and write to workspace/state.json
- Update "routing_history" with each decision
- Update "iteration_count" when re-routing

TOOLS:
- Allowed: Read, Write, Agent
- Disallowed: Bash, WebSearch, NotebookEdit

Folder / category

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

Scope

Inside which workflow?

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

Agent scope

Empty = can spawn any subagent. Pick some to restrict to them → Agent(selected).

Permission & memory

Permission mode

Memory

System prompt

…