worker

name	worker
description	Executes a single test case against the workspace and produces output. Use when running test cases, executing workspace tasks, or processing stage-specific work.
triggers	["run test case","execute workspace task","process stage","generate output"]

Overview

Execute a single test case by reading the relevant workspace sections, performing the required work, and producing structured output. Each worker runs with fresh context - no assumptions about prior runs.

When to Use

• Dispatched by orchestrator as part of a batch
• User asks to run a specific test case
• User asks to execute a workspace stage task

When Not to Use

• Validating outputs (use validation sub-skill)
• Fixing failed outputs (use fixer sub-skill)
• Planning workspace structure (use architecture sub-skill)

The Iron Law

NO SKIPPING TEST CASE STEPS NO MODIFYING WORKSPACE STRUCTURE NO CLAIMING DONE WITHOUT OUTPUT NO ASSUMING PRIOR CONTEXT

The Process

1. Read test case - Load the test case JSON from .agents/iteration/batch-<N>/<testCaseId>/ or orchestrator input
2. Load workspace context - Read SYSTEM.md and relevant stage CONTEXT.md files
3. Execute the task - Follow the test case input/expected instructions
4. Write output.md - Human-readable output in .agents/iteration/batch-<N>/<testCaseId>/output.md
5. Write report.json - Structured JSON with {skill, status, timestamp, testCaseId, batchId, findings, recommendations, metrics, nextSkill}
6. Dispatch validation - Signal that output is ready for validation

External Runner Contract

• Worker execution in autonomous iteration is orchestrated via --subagent-runner.
• Direct worker dispatch must provide --runner-command with placeholders {skill}, {workspace}, {batchId}, {testCaseId}.
• Runner output must be JSON and include report.json compatible fields:
  • skill, status, timestamp, findings, recommendations, metrics, nextSkill
• Missing/invalid runner output is a failure, not a simulated success path.
• Use .agents/iteration/runs/*.json telemetry to debug command rendering and runner payload issues.

Anti-Rationalization Table

Thought	Reality
"I already know what this stage does"	Read the CONTEXT.md. Assumptions cause failures.
"The output is good enough"	Good enough fails validation. Follow the test case exactly.
"I'll modify the workspace structure to make this easier"	Workers don't modify structure. That's the fixer's job.
"This test case is redundant"	Every test case exists for a reason. Execute it.
"I'll skip writing report.json"	Validation depends on report.json. It's mandatory.

Sub-Skill Dispatch

• After output complete -> validation sub-skill

Report Format

{
  "skill": "worker",
  "status": "passed|failed|escalated",
  "timestamp": "2026-04-08T00:00:00Z",
  "testCaseId": "tc-001",
  "batchId": 1,
  "findings": ["Output generated with required sections"],
  "recommendations": ["Proceed to validation"],
  "metrics": {
    "executionTimeMs": 120,
    "outputLength": 640
  },
  "nextSkill": "validation"
}