orien.runtime.yaml Reference

The runtime YAML defines how a Skill’s tools are loaded into the agent pipeline.

Complete Example


skill_id: my-strategy
version: 1.0.0
 
classification:
  category: trading          # trading | data_fetching | analysis | risk
  venue: hyperliquid         # Exchange or data source
  instrument_type: perpetual # perpetual | spot | futures
 
prompt_mapping:
  observe: |
    Analyze the current market for {symbol}.
    Use get_ticker for price data and get_klines for chart patterns.
  plan: |
    Based on the analysis, evaluate both bullish and bearish scenarios.
  risk: |
    Check position sizing and ensure risk limits are respected.
  execute: |
    Execute the recommended trade with proper stop-loss and take-profit.
 
tool_mapping:
  observe:
    - name: get_ticker
      entrypoint: scripts/market.py:get_ticker
      effect: read
    - name: get_klines
      entrypoint: scripts/market.py:get_klines
      effect: read
  execute:
    - name: place_order
      entrypoint: scripts/trading.py:place_order
      effect: write
 
tool_policy:
  observe:
    allowed_effects: [read]
  plan:
    allowed_effects: [read]
  risk:
    allowed_effects: [read]
  execute:
    allowed_effects: [read, write]
 
required_secrets:
  - EXCHANGE_API_KEY
  - EXCHANGE_API_SECRET
 
risk_controls:
  max_leverage: 5
  max_position_pct: 0.5

Field Reference

`classification`

Field	Values	Description
`category`	`trading`, `data_fetching`, `analysis`, `risk`	Primary purpose
`venue`	Exchange name	Target exchange
`instrument_type`	`perpetual`, `spot`, `futures`	Instrument type

`prompt_mapping`

Multi-line prompts for each phase. Used by prompt-type Skills and as context for the LLM in Python Skills.

`tool_mapping`

Maps tool functions to phases. Each tool has:

Field	Description
`name`	Tool name (used in LLM function calling)
`entrypoint`	`file:function` reference
`effect`	`read` or `write`

`tool_policy`

Restricts which effects are allowed per phase. The agent runtime enforces this — a write tool in the observe phase will be rejected.

`required_secrets`

Environment variables that must be set. The agent runtime validates these at startup and fails fast if missing.

`risk_controls`

Default risk parameters for this Skill. Users can override these during agent creation.