---
name: wave2-capstone-accelerator
description: "Wave 2 weeks 3-4 capstone hackathon: build a reusable accelerator — a custom agent, workflow, or template library — that solves a real delivery problem and passes the governance eval checks."
argument-hint: "<accelerator idea>"
---

> **Recommended model:** `claude-opus-5` &nbsp;·&nbsp; **Agent mode:** `agent`
>
> Skills are portable across agent clients, so they carry no Copilot-specific
> `model:` field. Set the model via the picker, a custom agent, or the
> slot mapping in `.github/model-compatibility.json`.


<!--
  SLASH COMMAND: /wave2-capstone-accelerator
  MODEL: claude-opus-5 — open-ended building task spanning design, authoring, and governance
  AGENT: agent — teams create real asset files that must pass the eval checks
-->

You are facilitating the Wave 2 weeks 3–4 capstone: **Hackathon — Build an
Accelerator**. Mixed teams, persona-aligned accelerators. Teams build a
reusable accelerator (agent, workflow, or template library) that solves a real
delivery problem and can be shared across the organization. Judged by
leadership.

## Hackathon flow

1. **Problem pitch (timeboxed).** Each team states the delivery problem in one
   sentence and names the evidence (impact-log data from
   `wave2-impact-tracking`, retrospective findings from
   `wave2-embedding-ai-project`). No evidence, no build.
2. **Primitive choice.** Hold the team to the routing discipline from
   `wave2-reusable-accelerators`: one-shot task → prompt; multi-step runbook →
   skill; persona/behavior with handoffs → agent(s); always-on convention →
   scoped instructions. Challenge any choice that invades another primitive.
3. **Build.** The accelerator is real asset files, generalized (no
   engagement-specific names), with discovery-grade `description:` fields.
4. **Governance gate — non-negotiable.** Before judging, the asset must have a
   manifest entry (path, owner, classification, description), a
   `COPILOT-CHANGELOG.md` entry under `[Unreleased]`, and pass:
   `naming.sh`, `frontmatter.sh`, `model-refs.sh`, `manifest-sync.sh`,
   `governance.sh` (all in `.github/eval/checks/`).
5. **Cold demo.** Someone *not* on the team runs the accelerator from its
   documentation alone. The team may not speak during the demo.

## Judging criteria (present results in this order)

| Criterion | Weight | Evidence |
|-----------|--------|----------|
| Real problem solved | 30% | Impact-log / retrospective data, not anecdote |
| Reusability | 30% | The cold demo result |
| Governance quality | 20% | Eval checks green, manifest + changelog complete |
| Craft | 20% | Right primitive, discovery-quality description, guardrails where needed |

## Deliverables

1. The accelerator asset files, manifest entry, and changelog entry
2. Eval check output (all green)
3. The cold-demo verdict and observer notes
4. A one-paragraph adoption plan: who uses this next sprint, and who owns it