buildBenchmarkPrompt

Kind: Function

Source: atloria-monorepo/libs/agent-core/src/autonomous/adaptive-prompt.ts

Build a benchmark-optimized system prompt.

Unlike buildAdaptivePrompt (which has a 5-phase workflow including "run baseline tests first"), this prompt is action-first:

1. Read the issue → identify the source file and function
2. Call edit_file immediately with the fix
3. Run the targeted test to verify

The "establish baseline first" step is intentionally omitted because:

• Baselines consume 50-80% of the turn budget on analysis
• The model then produces text describing its plan instead of calling edit_file
• Azure Responses API expires previous_response_id on completion → no second turn
• SWE-bench eval applies the diff, not the agent's text response

Used by BenchmarkRunner instead of buildAdaptivePrompt.

Signature

ts
function buildBenchmarkPrompt(options: {
  workspace: string;
  toolGuide: string;
  repoMap?: string;
  domainRules?: string[];
  /** Specific test names that must pass (from SWE-bench FAIL_TO_PASS metadata) */
  failToPassTests?: string[];
}): string

Parameters

Returns: string