ring/dev-team/commands/dev-cycle.md

---
name: ring:dev-cycle
description: Execute the development cycle for tasks in a markdown file
argument-hint: "[tasks-file] [options] [prompt]"
---

Execute the development cycle for tasks in a markdown file or from a direct prompt.

## Usage

```
/ring:dev-cycle [tasks-file] [options] [prompt]
/ring:dev-cycle [prompt]
```

## Arguments

| Argument | Required | Description |
|----------|----------|-------------|
| `tasks-file` | No* | Path to markdown file with tasks (e.g., `docs/tasks/sprint-001.md`) |
| `prompt` | No* | Direct instruction for what to implement (no quotes needed) |

*Either `tasks-file`, `prompt`, or `--resume` is required.

## Options

| Option | Description | Example |
|--------|-------------|---------|
| `--task ID` | Execute only specific task | `--task PROJ-123` |
| `--skip-gates` | Skip specific gates | `--skip-gates devops,review` |
| `--dry-run` | Validate tasks without executing | `--dry-run` |
| `--resume` | Resume interrupted cycle | `--resume` |

## Prompt Modes

### Direct Prompt (No Tasks File)

When a prompt is provided **without a tasks file**, ring:dev-cycle enters prompt-only mode:

1. **Analyzes the prompt** to understand what needs to be done
2. **Generates tasks internally** (similar to pre-dev but streamlined)
3. **Executes through 6 gates** as normal

```bash
# Direct prompt - no tasks file needed
/ring:dev-cycle Implement multi-tenant support for the API

# Another example
/ring:dev-cycle Add pagination to all list endpoints with cursor-based navigation
```

### With Tasks File + Additional Context

```bash
# Tasks file with additional context
/ring:dev-cycle docs/tasks/sprint-001.md Focus on error handling

# Tasks file with specific guidance
/ring:dev-cycle docs/tasks/sprint-001.md Prioritize security validations
```

**When to use direct prompt:**
- Quick implementations without full pre-dev planning
- Exploratory work where full planning is overkill
- When you know exactly what you want and want to skip task file creation

**When to use tasks file:**
- Complex multi-task features (use `/pre-dev-*` first)
- Refactoring work (use `/ring:dev-refactor` first)
- When you need documented task breakdown for tracking

### Prompt Behavior (with tasks file)

CANNOT override CRITICAL gates. Provides custom context to agents.

**Persistence:**
- **Stored:** `custom_prompt` field in `docs/ring:dev-cycle/current-cycle.json`
- **Resume:** Survives interrupts; editable by modifying state file before `--resume`
- **Reports:** Included in execution reports under "Custom Context Used"

**Propagation:** Prepended to all agent prompts in a dedicated `**CUSTOM CONTEXT (from user):**` section. Applies to all gates and all agent types.

**Constraints:**
- **Length:** Recommended 50-75 words (~300-400 chars); hard limit 500 chars (truncated with warning). For longer context, summarize or link to external docs.
- **Sanitization:** Whitespace trimmed, control chars stripped (except newlines), unicode normalized. Does not prevent semantic prompt-injection attempts.

**Gate Protection:**

CRITICAL: Gates 3, 4, 5 enforce mandatory requirements:
- MUST enforce 85% coverage (Gate 3 Testing)
- MUST dispatch all 5 reviewers (Gate 4 Review)
- MUST require user approval (Gate 5 Validation)

**Conflict Detection:**
- **Method:** Targeted pattern matching (e.g., `skip (gate|test|review)`, `bypass (threshold|check)`)
- **Output:** Warning to stderr + recorded in state file
- **Format:** `⚠️ IGNORED: Prompt matched pattern "[name]" at "[text]" — cannot override [requirement]`

See `ring:dev-cycle` skill's "Custom Prompt Injection" section for full validation logic.

**View/modify persisted prompt:**
```bash
jq '.custom_prompt' docs/ring:dev-cycle/current-cycle.json  # View
# Edit state file, then: /ring:dev-cycle --resume
```

## Examples

```bash
# Direct prompt - implement feature
/ring:dev-cycle Implement multi-tenant support with organization_id in all entities

# Direct prompt - another example
/ring:dev-cycle Add idempotency support to transaction create endpoint

# Execute all tasks from a file
/ring:dev-cycle docs/tasks/sprint-001.md

# Execute single task
/ring:dev-cycle docs/tasks/sprint-001.md --task AUTH-001

# Skip DevOps setup (infrastructure already exists)
/ring:dev-cycle docs/tasks/sprint-001.md --skip-gates devops

# Validate tasks without executing
/ring:dev-cycle docs/tasks/sprint-001.md --dry-run

# Resume interrupted cycle
/ring:dev-cycle --resume

# Tasks file with additional context
/ring:dev-cycle docs/tasks/sprint-001.md Prioritize error handling and use existing UserRepository interface
```

## Prerequisites

1. **Task file or prompt**: Either a markdown file generated by `/pre-dev-*` or `/ring:dev-refactor`, or a direct prompt
2. **Project standards** (optional but recommended): `docs/PROJECT_RULES.md` with project conventions
3. **PRD/TRD** (optional): If exists in `docs/pre-dev/{feature}/`, will be used in design phase

## Gates Executed

| Gate | Skill | Description |
|------|-------|-------------|
| 0 | `ring:dev-implementation` | Implement code (TDD) |
| 1 | `ring:dev-devops` | Create Docker/compose |
| 2 | `ring:dev-sre` | Observability (health checks, logging, tracing) |
| 3 | `ring:dev-unit-testing` | Unit tests with 85%+ coverage |
| 4 | `ring:dev-fuzz-testing` | Fuzz tests for edge cases |
| 5 | `ring:dev-property-testing` | Property-based tests for invariants |
| 6 | `ring:dev-integration-testing` | Integration tests with testcontainers |
| 7 | `ring:dev-chaos-testing` | Chaos tests for failure scenarios |
| 8 | `ring:requesting-code-review` | Code review (5 reviewers in parallel) |
| 9 | `ring:dev-validation` | Final validation |

**Note:** Tasks are loaded at initialization, not as a separate gate. All 10 gates are MANDATORY.

After all tasks: `ring:dev-feedback-loop` generates metrics report.

## Output

- **State file**: `docs/ring:dev-cycle/current-cycle.json` (feature) or `docs/ring:dev-refactor/current-cycle.json` (refactor)
- **Feedback report**: `docs/dev-team/feedback/cycle-YYYY-MM-DD.md`

## Related Commands

| Command | Description |
|---------|-------------|
| `/ring:dev-status` | Check current cycle status |
| `/ring:dev-cancel` | Cancel running cycle |
| `/ring:dev-report` | View feedback report |

---

## ⛔ MANDATORY: Load Full Skill

**This command MUST load the skill for complete workflow execution.**

```
Use Skill tool: ring:dev-cycle
```

The skill contains the complete 10-gate workflow with:
- Anti-rationalization tables
- Pressure resistance scenarios
- TDD sub-phases (Gate 0.1 RED → Gate 0.2 GREEN)
- Gate completion definitions
- Incremental compromise prevention
- State management schema

## Execution Context

Pass the following context to the skill:

| Parameter | Value |
|-----------|-------|
| `tasks-file` | First argument if it's a `.md` file path (e.g., `docs/tasks/sprint-001.md`) |
| `prompt` | Remaining text after tasks-file and options (direct instruction) |
| `--task` | If provided, filter to specific task ID |
| `--skip-gates` | If provided, list of gates to skip |
| `--dry-run` | If provided, validate only |
| `--resume` | If provided, resume from existing state file (ring:dev-cycle or ring:dev-refactor) |

**Argument Parsing:**
- If first argument ends with `.md` → treat as tasks-file
- If first argument is an option (`--*`) → no tasks-file
- Remaining non-option text → prompt

## Step 1: ASK EXECUTION MODE (MANDATORY)

**After loading skill and before executing gates, you MUST ask:**

```yaml
AskUserQuestion:
  questions:
    - question: "Select execution mode for this cycle"
      header: "Mode"
      options:
        - label: "Manual per subtask"
          description: "Checkpoint after each subtask"
        - label: "Manual per task"
          description: "Checkpoint after each task"
        - label: "Automatic"
          description: "No checkpoints, run all gates"
```

**Do not skip this.** User hints ≠ mode selection. Only explicit selection is valid.

## Quick Reference

See skill `ring:dev-cycle` for full details. Key rules:

- **all 6 gates execute** - Checkpoints affect pauses, not gates
- **Gates execute in order** - 0 → 1 → 2 → 3 → 4 → 5
- **Gate 4 requires all 5 reviewers** - 4/5 = FAIL
- **Coverage threshold** - 85% minimum, no exceptions
- **State persisted** - Can resume with `--resume` after any interruption