OpenAgentsControl/evals/agents/core/openagent/tests/contextscout-integration/01-known-context-direct-load.yaml

id: openagent-contextscout-known-context
name: "OpenAgent: Known Context - Should Load Directly (No ContextScout)"
description: |
  Tests that OpenAgent DOES NOT use ContextScout for standard tasks where
  the context path is well-known and hardcoded.
  
  This validates:
  - Agent loads .opencode/context/core/standards/code.md directly
  - Agent does NOT delegate to ContextScout (unnecessary overhead)
  - Fast execution without discovery phase
  
  Expected Behavior:
  - MUST load code.md before writing code
  - MUST NOT use task tool to call ContextScout
  - SHOULD complete in <30 seconds
  
  This test SHOULD FAIL if:
  - Agent uses ContextScout for known context (inefficient)
  - Agent doesn't load code.md at all (violates standards)
  - Agent takes >30s (discovery overhead detected)

category: developer

prompts:
  - text: |
      Write a new function called `fibonacci` that calculates the nth fibonacci number.
      Use TypeScript and follow the project's code standards.

approvalStrategy:
  type: auto-approve

behavior:
  mustUseTools:
    - read  # Must read code.md
  mustNotUseTools:
    - task  # Should NOT delegate to ContextScout
  minToolCalls: 2
  maxToolCalls: 10

expectedViolations:
  - rule: approval-gate
    shouldViolate: false
    severity: error

timeout: 30000  # Should be fast without discovery

tags:
  - openagent
  - contextscout-integration
  - known-context
  - performance
  - critical

# Expected outcome:
# - Agent reads .opencode/context/core/standards/code.md
# - Agent writes fibonacci function following standards
# - NO delegation to ContextScout
# - Completes in <30s