opencode.jsonc

{
  "$schema": "https://opencode.ai/config.json",
  "instructions": ["AGENTS.md"],

  "command": {
    "code-review": {
      "description": "Comprehensive multi-stage code review orchestration",
      "prompt": "You are a coordinator agent that executes a three-stage code review process:\n\n1. **Code Review**: Launch an `explore` subagent to analyze changes for quality, patterns, and best practices.\n2. **Evaluation**: After Stage 1 completes, launch a `general` subagent to critically assess the review findings against project goals (from AGENTS.md).\n3. **Conclusion**: After Stage 2 completes, launch a `general` subagent to synthesize findings and provide a final recommendation (Approve / Approve with changes / Request modifications / Reject).\n\nProcess **serially** (one stage at a time). Save each stage's output to the `review/` directory.\n\nAfter all stages complete, provide a summary with: overall recommendation, critical issues count, improvement suggestions count, and key decision factors."
    },
    "example-mapping": {
      "description": "Conduct an Example Mapping session to explore a feature before TDD",
      "prompt": "Conduct an interactive Example Mapping session with the user using four categories:\n\n- **Story (yellow)**: The feature or user story being explored\n- **Rules (blue)**: Business rules and acceptance criteria discovered\n- **Examples (green)**: Concrete examples that illustrate a rule — these become tests later\n- **Questions (red)**: Open questions and uncertainties needing clarification\n\n## CRITICAL: You are a FACILITATOR, not an inventor\n- NEVER invent a rule — Ask the user what the rules are\n- NEVER invent an example — Ask the user for concrete examples\n- NEVER assume boundary behavior — Ask what happens at the edges\n- NEVER fill in gaps yourself — If something is unclear, create a red card (question) and ASK the user\n\n## Mandatory Rules to Ask About\n1. **Zero / empty case**: What happens when there are no items / the input is empty?\n2. **Boundaries**: Is there a maximum? What happens beyond it?\n3. **Special cases**: Are there any special thresholds, bonuses, or exceptions?\n4. **Interactions**: Does this feature interact with or depend on other features?\n\n## Output\nWrite the result to a markdown file at the location the user specifies (default: `src/<feature-name>-example-mapping.md`).\n\nAfter writing, evaluate the health of the mapping:\n| Red cards (Questions) | Many -> Not ready |\n| Blue cards (Rules) | Many (>6) -> Too big |\n| Green cards (Examples) | Few -> Thin coverage |\n\nNext: If healthy, suggest using the red subagent to start TDD."
    },
    "tdd-auto": {
      "description": "Run autonomous TDD cycle (Red → Green → Refactor) for one test via subagents",
      "prompt": "The user wants to work through one TDD cycle autonomously. Execute sequentially:\n\n1. Launch `red` subagent to activate the next test and make it fail.\n2. After Red completes, launch `green` subagent to implement minimal code.\n3. After Green completes, launch `refactor` subagent to improve the code.\n\nEach subagent manages its own human checkpoint. Process serially — do NOT launch agents in parallel."
    }
  },

  "permission": {
    "bash": {
      "git *": "allow",
      "npm test*": "allow",
      "*": "ask"
    }
  }
}