team-project-scheduler — worked example of agentic-dev-starter

This repository is a worked example of the agentic-dev-starter methodology kit. It's the output you'd produce by following the kit's Protocol A on a real project — a small Node/TypeScript team scheduler — through five features and two dogfood validation runs.

It is not a product. It's a reference artifact: "here is what a spec-driven agentic dev project actually looks like, end to end."

Produced by unfolding the kit at commit 53cc0fb.

---

Why this repo exists

A team lead (call him Denis) evaluating agentic-dev-starter naturally asks: "can you show me this on a real project?" Reading docs/FAQ.md and docs/METHODOLOGY.md in the kit explains the rules. This repo demonstrates them:

• A full 7-document Blueprint Package authored with the kit's interview
• Five features specified, decomposed, tested, implemented, and merged
• Two dogfood sessions where the kit was used to build features in the kit's own author's hands — stress-testing the enforcement layer on real work
• A real Construction Site (CS-002) left intentionally open to show how the methodology catches scope deferral instead of hiding it

What the project does

A scheduler for small teams:

• Unified weekly calendar view of everyone's tasks (Scenario 1)
• Slack notifications on task assignment (Scenario 2)
• Explicit task priorities (Scenario 3)

Stack: Node 20 · TypeScript 5.5 · Vitest · ESM. No UI yet — the 006-calendar-ui feature is committed-to as a sibling of 005, and its absence is the reason CS-002 is open. Build is server-only; 26/26 unit + integration tests pass; 81% coverage on branch-diff scope.

This is deliberately a small project. The methodology's value isn't proportional to project size — it's proportional to how well a team can see their work as they go. A tiny project with full gates visible beats a large project where you have to imagine the scaffolding.

Reading order (for a team lead evaluating the kit)

1. CONSTITUTION.md — hard rules, Articles 1–10. Article 10 is customized for this project (Node/TS, single-lead team, hobbyist-scale operations).
2. docs/00-CORE-PRINCIPLES.md — what problem, for whom, the 3 principles that generate the scenarios.
3. docs/01-SYSTEM-INTENT.md — 3 Experience Fidelity Scenarios with ≥3 negative assertions each, filmable success criteria, the Scenario Validation Matrix with bidirectional assertion↔task linkage, and depth classification.
4. docs/02-ARCHITECTURE.md — module boundaries, API surface, data flows, and an Architecture Impact Assessment per scenario.
5. docs/03-EXECUTION-CONTEXT.md — pinned stack versions, testing philosophy, error handling.
6. docs/04-COORDINATION-HINTS.md — phase ordering with depth-tagged done criteria, Production Threshold, 10 non-goals.
7. docs/05-CONSTRUCTION-SITES.md — the living debt tracker. Currently: CS-001 RESOLVED (cross-feature shared-model merge conflict, addressed by installing the dna:cross-checker subagent). CS-002 OPEN — the reason 005 didn't ship with a UI.
8. RETROSPECTIVE_INDEX.md — one retrospective per feature, in merge order. The honest chronology.
9. dogfood-evidence/ — raw session notes from the two dogfood runs. This is where you see the gates firing, the subagents returning verdicts, and the 9 rough edges the kit's own author found by using the kit.

What to observe

Artifact	What it demonstrates
docs/01-SYSTEM-INTENT.md Scenario 1	Filmable success statement + 3 negative assertions + Validation Matrix with task IDs pointing back to spec assertions
docs/02-ARCHITECTURE.md Impact Assessments	Every scenario forces an architectural impact paragraph — prevents drift at design time
docs/05-CONSTRUCTION-SITES.md CS-002	A [D] scenario legitimately shipped at partial depth under Article 5 (out-of-scope + sibling feature + logged entry + phase not closed)
specs/004-task-status-transitions/retrospective.md	First full 12-step dogfood. Read what the kit felt like to use rather than what it says
specs/005-calendar-week-grid/retrospective.md	Second dogfood, [D] depth. Nine rough edges named and catalogued
dogfood-evidence/DOGFOOD-NOTES-2026-04-23.md	The judgmental subagent (dna:spec-validator) correctly distinguishing legitimate scoping from production-threshold drift
.github/workflows/dna.yml	CI enforcement for dna-decompose + dna-verify + dna-spec-validate (no headless LLM path — those are dev-time gates)
.claude/skills/dna-*/	Six executable enforcement skills, copied directly from the kit by Protocol A
.claude/agents/dna-*	Five subagents, each a fresh-context judge

What this repo is NOT

• Not a product — don't try to deploy it or rely on it. The scheduler exists as connective tissue for demonstrating the methodology.
• Not a starter template — for that, go to agentic-dev-starter and run Protocol A in your own project.
• Not maintained in lockstep with the kit — this is a snapshot at kit commit 53cc0fb. Future kit releases may introduce changes this repo doesn't reflect.
• Not feature-complete — CS-002 is open intentionally. If it bothers you to see an open Construction Site in a "demo," that's the point.

Want the kit itself?

github.com/albertdobmeyer/agentic-dev-starter

Clone it, point Claude Code at it, say "set up a new project for my team." The agent will interview you and produce a repo like this one, shaped around your project's vision.

---

Licensed under CC BY-SA 4.0. Created by Albert Dobmeyer & Claude (Anthropic) — AKD AUTOMATION SOLUTIONS.