Workflow State Skill

Purpose

Manages sdd/workflows/<id>-<name>/ state - tracking where each item is in the solicitation → review → approval → implementation lifecycle.

This is process state management, not project task management.

Core Principle: Zero Session Context

ALL workflow state is persisted in sdd/ files. A new session must be able to resume with ZERO knowledge of what happened before:

• No conversation history
• No in-memory state
• No assumptions

Read the files, know the state. This enables aggressive context compaction and allows users to clear sessions freely.

Directory Structure

sdd/
├── sdd-settings.yaml
├── archive/
│   ├── external-specs/         # External specs archived here (read-only)
│   │   └── 20260205-1430-feature-spec.md  # yyyymmdd-HHmm-lowercased-filename.md
│   ├── revised-specs/          # Specs removed during decomposition revision
│   │   └── 20260205-1430-a1b2c3-user-auth-03-password-reset/
│   │       └── SPEC.md
│   └── workflow-regressions/   # Work archived during phase regression
│       └── 20260205-1430-user-auth-1-impl/
│           ├── changes.patch   # Git patch for committed changes
│           ├── stash.patch     # Git stash for uncommitted changes
│           ├── src/            # Implementation files
│           └── metadata.yaml   # Regression context
└── workflows/                  # Multiple concurrent workflows supported
    ├── a1b2c3-user-auth/       # One workflow (user A, branch feature-x)
    │   ├── workflow.yaml       # This workflow's state
    │   └── drafts/
    │       ├── 01-user-management/
    │       │   ├── context.md
    │       │   ├── 01-api-contracts/
    │       │   │   └── context.md
    │       │   └── 02-auth-service/
    │       │       └── context.md
    │       └── 02-notifications/
    │           └── context.md
    └── x7y8z9-notifications/   # Another workflow (user B, branch feature-y)
        ├── workflow.yaml
        └── drafts/...

Status Progression (Four-Field Model)

The workflow uses four separate status fields for granular tracking:

Spec Phase

1. spec_status: pending → in_progress → ready_for_review → approved

Plan Phase (only after ALL specs approved)

2. plan_status: pending → in_progress → approved

Implementation Phase (only after ALL plans approved)

3. impl_status: pending → in_progress → complete

Review Phase (after implementation)

4. review_status: pending → ready_for_review → approved

Key Difference from Previous Design:

The old design had a single status field with immediate spec_approved → plan_review transition. The new design:

1. spec_status: approved IS a resting state - Phase gating requires ALL specs approved before ANY planning starts
2. Plan phase is separate - Only begins after checkpoint "All specs approved"
3. Review phase is explicit - User must approve implementation, not just verify tests pass

Phase Gating Rules

Gate	Condition
Start planning	ALL items have spec_status: approved
Start implementing	ALL items have plan_status: approved
Complete workflow	ALL items have review_status: approved
Approve spec	No OPEN questions in Requirements Discovery section
Approve spec	No dependencies with spec_status: needs_rereview

Resource Files

For detailed guidance, read these on-demand:

• internal-api.md — All 15 operations with input/output schemas
• workflow-yaml-schema.md — Full schema documentation with all fields
• recovery.md — Checkpoint triggers, recovery scenarios, regression handling

Input

Schema: schemas/input.schema.json

Accepts operation type and operation-specific parameters for workflow lifecycle management.

Output

Schema: schemas/output.schema.json

Returns workflow ID, current phase, and progress tracking.

Workflow ID Generation

Workflow IDs are short, unique identifiers:

1. Generate 6 random alphanumeric characters (a-z, 0-9)
2. Check for collision with existing workflows
3. If collision, regenerate
4. Format: a1b2c3, x7y8z9, etc.

Change ID Format

Change IDs are derived from the workflow name:

• Format: <name>-<seq> (e.g., user-auth-1, user-auth-2)
• name: The workflow's user-chosen name
• seq: Sequence number within workflow (1, 2, 3, ...)
• Unique across concurrent workflows (different workflows = different name prefix)

Epic Handling

• Each epic is an item with type: epic and children array
• Children are features within the epic
• Epic status tracks overall progress:
  • pending until first child starts
  • complete when all children complete
• Epic itself doesn't get a change_id - only leaf features get change_ids
• Epic folder in changes/ contains child change folders

Cleanup Behavior

• When item moves from drafts/ to changes/:
  • Keep context.md in drafts for reference
  • Delete solicitation-workflow.yaml
• When item is marked complete:
  • Remove from workflow.yaml items array
• When all items complete:
  • Delete entire sdd/workflows/<id>-<name>/ directory including workflow.yaml
• Completed items remain in changes/ permanently (that's the source of truth)

INDEX.md Handling

• changes/INDEX.md is updated when items move to changes/ directory
• No separate INDEX.md in sdd/workflows/ - workflow.yaml is the source of truth
• workflow_state.ready_for_review() updates changes/INDEX.md with new entry

Consumers

• /sdd orchestrator — creates workflows and manages lifecycle
• spec-solicitation skill — reads context, saves specs
• change-creation skill — saves plans
• external-spec-integration skill — creates workflows and items from decomposition