AGENT SKILL

project-launch-checklist

Build and review two-tier production launch checklists — generic + framework-specific — and author vault reference checklists (ai-checklist family). Simplify verbose reference manuals into tickable checklists (ponytail pattern).

License 许可
Unlicense
最近更新
2026-11-03
Stars
★ 1
Forks
0

01 前言 Front Matter

name
project-launch-checklist
description
Build and review two-tier production launch checklists — generic + framework-specific — and author vault reference checklists (ai-checklist family). Simplify verbose reference manuals into tickable checklists (ponytail pattern).

02 资源文件 1

03 SKILL.md 正文

Project Launch Checklist — Two-Tier System

Build production-ready launch checklists for any tech stack. Pattern: generic tick-first checklists + framework-specific companions + reference manuals.

The Two-Tier Pattern

Three file types per domain, living in the same folder:

domain-checklist/
├── {Domain} Launch.md           ← Generic. Tick FIRST. ~40-50 items, framework-agnostic.
├── framework-vX-api.md          ← Framework companion. Tick AFTER generic. ~70-85 items.
├── framework-reference.md       ← Reference manual. Deep dives, tutorials. Not ticked.
└── (repeat for each framework)

Tier 1: Generic Launch Checklist

  • 6-8 sections, 5-7 one-line items each
  • Target: ~45 items total
  • Every item: - [ ] Description → [[vault-note]] — links to knowledge vault, no explanations embedded
  • Zero code examples. Zero version trivia. Zero tutorials.
  • Sections: Security → API/State → Database/Data → Resilience/Performance → Observability → Testing → Deployment → (+ Routing for frontend)
  • References framework companions: "For {framework} specifics, see [[framework-vX]]"
  • References dependent checklists: "Assumes [[Related Launch]] is already done"

Tier 2: Framework Companion

  • "Tick [[{Domain} Launch]] first" in the header
  • Framework-specific only: setup, routing, DI, middleware, ORM, testing tools
  • ~70-85 items. Still tickable. Still no tutorials.
  • Sections mirror generic but with framework details: Project Setup → App Structure → Middleware/Routing → Auth → Config/Secrets → DB/State → Testing → Observability → Build/Deploy → Quick Sanity Check
  • Quick Sanity Check at the end: 8-10 items verifiable in 2 minutes

Tier 3: Reference Manuals (Keep Separately)

  • Can be long (100-250 lines), detailed, with code examples
  • Version-specific (Spring Boot 4.x, Fiber v3, React 19+, Angular 17+)
  • Contains tutorials, migration guides, option comparisons
  • NOT ticked before launch — consulted when a launch checklist item fails
  • Named without "Launch" suffix: spring-boot-api.md, react-js.md

Ponytail Filter

When reviewing existing verbose checklists:

  1. Strip tutorials, code examples, and version-specific trivia
  2. Keep only tick boxes (- [ ]) with one-line descriptions
  3. Link to vault notes (→ [[vault-note]]) instead of embedding explanations
  4. Target ~40-50 items per file
  5. Keep original as deep reference; create -v2.md or rename to {Domain} Launch.md

If a checklist item needs > 1 line of explanation, it belongs in the reference manual, not the launch checklist.

Reference Checklists (ai-checklist family — a second house style)

checklist/ai-checklist/ is not the tick-first launch family. Its files (ai.md, claude-driven.md, general-agents-driven.md, plus topic files added since) are reference checklists: deeper, explanation-carrying, still tickable. Same vault, different style — match the family you are writing into.

House style of the reference family:

  • Filename: lowercase-hyphen.md (loop-engineering.md, never Loop Engineering.md)
  • H1: # <Topic> Checklist
  • Header blockquote: one-line purpose → Siblings: cross-links → Last updated: YYYY-MM-DD (new file — <why>); bump the date on every later edit
  • Numbered sections (## 1. …), items as - [ ] **Bold lead** — one-sentence explanation. with inline → [[link]] §n pointers
  • House-standard extras where they fit: a Scoping tier table adapted from the tier model in [[ai]] / [[general-agents-driven]], a Quick Pre-Flight minimum-bar list, and a Sources footer attributing each reference (source — title (date): what it contributed)
  • ⚠️ Emerging or contested topics: draw on several primary sources (canonical vendor engineering posts + practitioner write-ups), state plainly that the terminology is unsettled, and never present a new term as a settled standard

Authoring workflow:

  1. Read 1–2 sibling files in the target folder first and match the living style — styles drift; the most recent file in the folder beats any written convention
  2. Cross-link consistently: siblings by [[lowercase-hyphen]]; cross-vault notes by path-style with alias ([[career-path/…/03_Agent_Loops_and_Orchestration|Agent Loops and Orchestration]]); never spaces in backlinks
  3. Verify before reporting done: every [[wikilink]] resolves, no ambiguous basenames → obsidian-vault-maintenance → scripts/verify_files.py
  4. checklist/Overview.md indexes only the original "improved" launch set — it does not list the ai-checklist family. Do not silently restructure it; offer the index update

"Repo or checklist?" — when the user asks whether a topic deserves a repository, default to a vault checklist. A repository earns its existence only once executable artifacts exist (harness skeleton, example topologies, eval scripts); then the checklist becomes the repo's docs and the vault keeps a pointer.

Cross-Referencing Rules

  • Generic → Framework: "For {framework} details, see [[framework-vX-api]]"
  • Framework → Generic: "Tick [[{Domain} Launch]] first"
  • Generic ↔ Vaults: Every item links → [[vault-note]] for deep knowledge

Common Missing Sections

When reviewing checklists, always check for these commonly missing sections:

  • Routing (frontend): file-based routing, dynamic routes, navigation guards, 404
  • Form validation: Zod/schema + server-side re-validation (client UX, server security)
  • OpenAPI/Swagger (backend): auto-generated docs, annotations
  • Secrets management: never in source code, Vault/K8s/env vars
  • Graceful shutdown: drain connections, close DB on SIGTERM

Auditing Existing Checklists (Domain Specialist Review)

When a user asks to "audit" or "review" an existing checklist for production readiness:

1. Read ALL files in the checklist folder first

Checklists are usually a collection: a general file + engine/topic-specific companions. Read every file in the folder before auditing — cross-file consistency matters (shared tier matrix format, shared wikilink targets, consistent section depth).

2. Check for mandatory cross-cutting sections

Every production-grade checklist (database, AI, infrastructure) should have these cross-cutting sections. If missing, flag as 🔴 critical:

  • Privacy & Data Protection — PII inventory, right to erasure, retention, anonymization, data subject requests
  • Data Quality — validation rules, profiling, drift monitoring, integrity checks
  • Maintenance & Operations — routine maintenance, vacuum/compaction, statistics freshness, deadlock detection

These are the three sections most commonly missing from checklists written by application engineers — they think about "does it work" not "is the data correct/safe/maintainable over time."

3. Engine-specific gap analysis

Each engine has unique operational concerns that a general checklist can't cover. When auditing an engine-specific checklist, verify the engine's signature features are present (not just general DB principles). See references/database-checklist-audit-domains.md for the full per-engine gap list (PostgreSQL, MongoDB, Valkey, Elasticsearch, ClickHouse, SQLite, Cassandra, plus AI/LLM-specific sections).

4. Tier matrix synchronization

When you add or expand sections in a checklist:

  • The tier scoping matrix table must grow correspondingly — one row per section
  • The Quick Sanity Check must add launch-critical items from new sections
  • Re-verify: does every section in the checklist body have a row in the tier matrix?

5. Honest scope warnings in tier matrices

Add explicit warnings when a technology doesn't fit certain tiers:

  • SQLite: "⚠️ Typically not SQLite territory at Tier 5+ — consider migrating to PostgreSQL"
  • Cassandra: "⚠️ Overkill below Tier 4 — choose when you need >10K writes/sec or multi-DC active-active" This prevents the "golden hammer" problem where engineers scale a technology past its design intent.

6. Section renumbering

When adding sections to the middle of a checklist, renumber ALL subsequent sections and update ALL internal cross-references (e.g., "see §4"). A full write_file rewrite is cleaner than multiple fragile patch calls when renumbering changes most of the file.

7. Update cross-references in companion files

When a general checklist gains new engine companions (e.g., database.md now references elasticsearch.md), update the general file's header to list all companions. Keep wikilinks consistent across the collection.

8. "Last updated" date

Bump the date in the header blockquote on every file you modify.

Framework Picks (Ponytail Default)

When user asks "what framework for X?" without experience:

  • Go → Fiber v3 (simpler than stdlib, lighter than Echo, Axum for Rust fans)
  • Rust → Axum (tokio-native, no macros, great docs. NOT Actix/Rocket/Warp for beginners)
  • Node → NestJS (opinionated, DI, TypeScript-first, Express/Fastify)
  • Frontend state → TanStack Query for server state, signals/stores for client state. No Redux.

04 同仓库更多 Skills