/approve

description: "Approve the current PRD and begin parallel story execution. Analyzes dependencies, creates execution batches, launches background Task agents for each story, and monitors progress. Usage: /plan-cascade:approve [--flow <quick|standard|full>] [--tdd <off|on|auto>] [--confirm] [--no-confirm] [--agent <name>] [--impl-agent <name>] [--retry-agent <name>] [--no-verify] [--verify-agent <name>] [--no-review] [--no-fallback] [--auto-run]"

Hybrid Ralph - Approve PRD and Execute

You are approving the PRD and starting parallel execution of user stories.

Execution Flow Parameters

This command accepts flow control parameters that affect quality gates and execution:

Parameter Priority

Parameters can be specified in three places. When the same parameter is defined in multiple sources, the following priority order applies:

1. **Command-line flags** (highest priority)

• Example: `/plan-cascade:approve --flow full`
• Overrides all other sources

2. **PRD configuration** (prd.json)

• `flow_config.level`, `tdd_config.mode`, `execution_config.require_batch_confirm`
• Used when no command-line flag is provided

3. **Default values** (lowest priority)

• Used when no command-line flag or PRD config is present

**Example Priority Resolution:**

# Scenario 1: Command-line overrides PRD
prd.json contains: flow_config.level = "standard"
You run: /plan-cascade:approve --flow full
Result: FLOW = "full" (command-line wins)

# Scenario 2: PRD config used when no command-line flag
prd.json contains: tdd_config.mode = "on"
You run: /plan-cascade:approve
Result: TDD = "on" (from PRD)

# Scenario 3: Default used when neither specified
prd.json has no tdd_config
You run: /plan-cascade:approve
Result: TDD = "auto" (default)

**Debugging Tip:** The command displays resolved parameter values at Step 2.1, showing which source each parameter came from.

`--flow <quick|standard|full>`

Override the execution flow depth. This controls quality gate strictness.

| Flow | Gate Mode | AI Verification | Code Review | Test Enforcement | |------|-----------|-----------------|-------------|------------------| | `quick` | soft (warnings) | **disabled** | no | no | | `standard` | soft (warnings) | enabled | no | no | | `full` | **hard (blocking)** | enabled | **required** | **required** |

**FULL Flow Enforcement:**

• Quality gates BLOCK execution on failure (not just warn)
• Code review is REQUIRED after each story
• Test file changes are REQUIRED alongside code changes
• Batch confirmation is shown before each batch

`--tdd <off|on|auto>`

Control Test-Driven Development mode for story execution.

| Mode | Description | |------|-------------| | `off` | TDD disabled, no compliance checks | | `on` | TDD enabled with prompts and compliance gate | | `auto` | Automatically enable TDD for high-risk stories (default) |

`--confirm`

Require explicit user confirmation before starting each batch execution.

`--no-confirm`

Explicitly disable batch confirmation, even if FULL flow would normally require it.

This is useful for:

• CI/CD pipelines where interactive confirmation is not possible
• Automated testing environments
• When you want strict quality gates but uninterrupted execution

**Note**: `--no-confirm` only affects batch-level confirmation. It does NOT disable quality gates (verification, code review, TDD compliance) - those still run and can block on failures in FULL flow.

**Precedence**: `--no-confirm` overrides `--confirm`, PRD config, and FULL flow's default confirmation requirement.

Execution Modes

The approve command supports three execution modes:

Mode 1: Auto Mode (Default in traditional flow)

• Automatically progresses through batches
• Pauses only on errors requiring manual intervention
• Uses Claude-based task agents for story execution

Mode 2: Manual Mode

• Requires user approval before each batch
• Full control and review capability
• Best for high-risk or learning scenarios

Mode 3: Full Auto Mode (Recommended)

• **Python-based execution with automatic retry**
• Quality gates run automatically with auto-retry on failures
• Up to 3 retry attempts per story with exponential backoff
• Failure context injected into retry prompts
• Best for CI/CD and unattended execution

**Full Auto Mode Features:** | Feature | Description | |---------|-------------| | Auto Retry | Failed stories automatically retry up to 3 times | | Failure Context | Error details and suggested fixes added to retry prompts | | Exponential Backoff | 5s → 10s → 20s → ... delays between retries | | Agent Switching | Optional: switch to `--retry-agent` on failures | | State Persistence | Progress saved to `.iteration-state.json` for recovery |

To use Full Auto mode, select option `[3]` when prompted for execution mode, or run:

uv run python scripts/auto-execute.py --prd prd.json --flow full --tdd on

Path Storage Modes

This command works with both new and legacy path storage modes:

New Mode (Default)

• PRD file: In worktree directory or `~/.plan-cascade/<project-id>/prd.json`
• State files: `~/.plan-cascade/<project-id>/.state/`
• Agent outputs: In worktree or project root `.agent-outputs/`

Legacy Mode

• PRD file: In worktree or project root `prd.json`
• State files: In project root
• Agent outputs: In project root `.agent-outputs/`

User-visible files (`progress.txt`, `findings.md`) always remain in the working directory.

Multi-Agent Collaboration

This command supports multiple AI agents for story execution. The system automatically selects the best agent based on:

1. Command-line parameters (highest priority) 2. Story-level `agent` field in PRD 3. Story type inference (bugfix→codex, refactor→aider) 4. Phase defaults from agents.json 5. Fallback to claude-code (always available)

Supported Agents

| Agent | Type | Best For | |-------|------|----------| | `claude-code` | task-tool | General purpose (default, always available) | | `