id: openrouter-context-loading-exec
name: "OpenRouter Variant: Context Loading Execution Test"
description: |
  EXECUTION TEST - Validates agent actually loads context files before writing code.
  
  Expected workflow:
  1. User requests code to be written
  2. Agent proposes plan (approval gate)
  3. User approves
  4. Agent loads .opencode/context/core/standards/code.md (using read tool)
  5. Agent writes the file
  
  This test validates:
  - Agent uses read tool to load context file
  - Context file path matches .opencode/context/core/standards/code.md
  - Context is loaded BEFORE write tool is used

category: developer
agent: openagent
model: x-ai/grok-beta

prompts:
  - text: |
      Create a simple utility function in evals/test_tmp/openrouter-utils.js 
      that exports a function called greet(name) which returns "Hello, {name}!".
      Keep it simple - just the function.
  
  - text: |
      Yes, proceed with the plan.
    delayMs: 3000

behavior:
  mustUseTools:
    - read      # Must load context
    - write     # Must write file
  requiresApproval: true
  minToolCalls: 2

expectedViolations:
  - rule: approval-gate
    shouldViolate: false
    severity: error
    description: Should request approval before writing
  
  - rule: context-loading
    shouldViolate: false
    severity: error
    description: Must load .opencode/context/core/standards/code.md before writing code

approvalStrategy:
  type: auto-approve

timeout: 90000

tags:
  - execution
  - openrouter
  - context-loading
  - code-standards
  - tool-validation