amadeus-outcomes-pack

AI-DLC Outcomes Pack

Purpose

Produce a single handover document at the close of a workflow that gives the team everything they need to own, operate, and keep building the delivered system — without re-running the workflow to recover context.

Classification

Read-only with respect to workflow state. This skill never advances the stage pointer and never emits an audit event. It does write one report artefact (OUTCOMES.md at the workspace root) — that is its output. It writes nothing else.

The counting rule

Stage tallies, per-phase rollup, memory-entry counts, and learnings captured come from the tool, not from eyeballing the artefact tree:

bun {{HARNESS_DIR}}/tools/amadeus-runtime.ts summary --json

Section content (what was built, setup steps, decisions) is yours to synthesise from the artefacts and the delivered code. Any count that appears in the pack must trace to the tool's output.

Steps

Step 1: Read the aggregates

Run bun {{HARNESS_DIR}}/tools/amadeus-runtime.ts summary --json.

If it exits non-zero (no runtime-graph.json yet), print:

No session data yet — an outcomes pack is generated at the close of a
workflow. Run /amadeus to completion first.

and STOP. Otherwise keep the parsed JSON for the count fields below.

Step 2: Read the content sources

• All artefacts under amadeus-docs/<phase>/<stage>/ — requirements, design decisions, NFRs, infrastructure.
• The delivered application and infrastructure code at the workspace root.

Step 3: Write OUTCOMES.md

Write OUTCOMES.md at the workspace root (not inside amadeus-docs/):

# Outcomes Pack
**Scope**: {summary.scope}
**Stages delivered**: {summary.stages.approved} approved / {summary.stages.total} total
**Duration**: {summary.duration_minutes} min

## 1. What Was Built
- Project name and description (from requirements)
- Scope the workflow ran at
- Units of work delivered and what each contains
- Key architectural decisions and why (from design artefacts)
- Tech stack with version pins

## 2. Repository Structure
- Annotated directory tree of the delivered code
- What lives where and why

## 3. Setup Guide
- Prerequisites (runtimes, tools, cloud CLI versions)
- Local development setup, step by step
- Required environment variables
- How to run tests

## 4. Build and Deploy
- Build steps
- Full test-suite run
- Infrastructure deployment (from Build and Test artefacts)
- IaC deployment commands with expected outputs, if generated

## 5. Architecture Decisions
- Every significant decision made during the workflow
- Alternatives considered and why rejected
- Constraints that shaped the design (from rules and practices)

## 6. What to Commit vs Archive
| Artifact | Action | Destination |
|----------|--------|-------------|
| `decisions.md` (per stage) | Commit | `docs/decisions/` |
| Architecture summary (1 page) | Write + commit | `docs/architecture.md` |
| NFR summary table | Write + commit | `docs/nfr-summary.md` |
| `audit.md` | Archive — do NOT commit to app repo | Compliance archive |
| Stage question files | Discard | — |
| `amadeus-state.md` | Discard | — |
| Application / infrastructure code | Already committed | — |

## 7. Workflow Footprint
- Stages: {summary.stages.approved} approved, {summary.stages.failed} failed, {summary.stages.pending} pending
- Memory entries captured: {summary.memory.total}
  ({summary.memory.interpretations} interpretations, {summary.memory.deviations} deviations, {summary.memory.tradeoffs} trade-offs, {summary.memory.open_questions} open questions)
- Learnings captured: {summary.learnings.from_orchestrator} from orchestrator, {summary.learnings.from_user_addition} from user additions

## 8. Known Limitations and What to Tackle Next
- Scope items explicitly deferred during the workflow
  (cross-reference the {summary.memory.open_questions} open questions above)
- Technical debt identified but not resolved
- Recommended next steps