Idd Orchestration

---
name: IDD Orchestration
description: This skill should be used when the user asks to "start IDD", "use intent-driven development", "set up IDD workflow", "define a product", "write intentions", "create expectations", "author a spec", "review a spec", "run the IDD process", or invokes any /idd:* command. Orchestrates the Intent-Driven Development workflow for AI-augmented teams.
---

# IDD Orchestration

Intent-Driven Development decomposes purpose into four levels — Product, Intention, Expectation, and Spec — giving developers and AI agents the context they need to make implementation decisions independently.

## Workflow Phases

| Phase | Command | Agent | Artifact | Output Path |
|-------|---------|-------|----------|-------------|
| 1. Interview | `/idd-framework:interview` | product-interviewer | Product definition | `docs/products/` |
| 2. Intentions | `/idd-framework:define-intentions` | intention-author | Intention artifacts | `docs/intentions/` |
| 3. Expectations | `/idd-framework:define-expectations` | expectation-author | Expectation artifacts | `docs/expectations/` |
| 4. Spec | `/idd-framework:write-spec` | spec-author | Spec artifact | `docs/specs/` |
| 5. Tech Review | `/idd-framework:tech-review` | tech-lead-reviewer | Review annotations | `docs/specs/` (updates) |
| 6. Validation | `/idd-framework:review-spec` | spec-reviewer | Validation report | `docs/reviews/` |

**Accelerated workflows:**
- `/idd-framework:define-outcomes` — Combines Intentions + Expectations in one session
- `/idd-framework:quick-spec` — Produces Intentions + Expectations + Spec in one session
- `/idd-framework:deep-review` — Multi-perspective review using Agent Teams

## Entry Points

Users can enter the workflow at any phase:

- **Full pipeline:** Start with `/idd-framework:interview`
- **Already have a product?** Start at `/idd-framework:define-intentions`
- **Have intentions?** Start at `/idd-framework:define-expectations`
- **Have expectations?** Start at `/idd-framework:write-spec`
- **Have a spec to review?** Start at `/idd-framework:tech-review`
- **Have AI output to validate?** Start at `/idd-framework:review-spec`

## Agents

| Agent | Color | Model | Role |
|-------|-------|-------|------|
| `idd-product-interviewer` | blue | **haiku** | Interviews stakeholder to capture Product artifact |
| `idd-intention-author` | green | **haiku** | Guides Product Owner to decompose Product into Intentions |
| `idd-expectation-author` | yellow | **haiku** | Defines verifiable Expectations with edge cases |
| `idd-outcome-author` | green | **sonnet** | Defines Intentions + Expectations in one session |
| `idd-spec-author` | cyan | **sonnet** | Creates AI-ready Specs with all 5 mandatory blocks |
| `idd-quick-spec-author` | cyan | **sonnet** | Produces INT + EXP + SPEC in one session |
| `idd-spec-reviewer` | red | **sonnet** | Validates AI output against Spec criteria |
| `idd-tech-lead-reviewer` | magenta | **opus** | Reviews Specs for architectural feasibility |
| `idd-deep-review-lead` | magenta | **opus** | Multi-perspective Spec review (parallel) |

## CRITICAL: Model Dispatch Rule

**When dispatching any `idd-*` subagent (via the Agent or Task tool), you MUST explicitly pass the `model` parameter in the tool call, matching the agent's tier above.** Do NOT rely on the subagent's frontmatter `model:` field alone — in practice, orchestrators frequently dispatch subagents with the parent session's model unless the model is set explicitly at dispatch time. This defeats the plugin's cost-optimization design and can turn a cheap Haiku interview into an expensive Opus run.

Correct pattern (pseudocode):
```
Agent({
  subagent_type: "idd-intention-author",
  model: "haiku",                         // REQUIRED — match the table above
  prompt: "...",
})
```

If you are invoking one of these subagents and are unsure of the tier, read the agent's frontmatter file at `plugin/agents/<name>.md` and use the `model:` value there as the source of truth. The per-command `.md` files under `plugin/commands/` also state the required model inline.