{"id":48451830,"url":"https://github.com/all-the-vibes/atv-starterkit","last_synced_at":"2026-04-27T05:01:24.175Z","repository":{"id":343814201,"uuid":"1179178348","full_name":"All-The-Vibes/ATV-StarterKit","owner":"All-The-Vibes","description":"A one-click installer that scaffolds a complete GitHub Copilot agentic coding environment into any project","archived":false,"fork":false,"pushed_at":"2026-04-24T17:18:12.000Z","size":16289,"stargazers_count":40,"open_issues_count":5,"forks_count":8,"subscribers_count":1,"default_branch":"main","last_synced_at":"2026-04-24T17:20:41.836Z","etag":null,"topics":["agentic-coding","developer-tools","github-copilot","starter-kit"],"latest_commit_sha":null,"homepage":null,"language":"Go","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/All-The-Vibes.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-03-11T19:10:22.000Z","updated_at":"2026-04-24T16:52:04.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/All-The-Vibes/ATV-StarterKit","commit_stats":null,"previous_names":["all-the-vibes/atv-starterkit"],"tags_count":13,"template":false,"template_full_name":null,"purl":"pkg:github/All-The-Vibes/ATV-StarterKit","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/All-The-Vibes%2FATV-StarterKit","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/All-The-Vibes%2FATV-StarterKit/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/All-The-Vibes%2FATV-StarterKit/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/All-The-Vibes%2FATV-StarterKit/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/All-The-Vibes","download_url":"https://codeload.github.com/All-The-Vibes/ATV-StarterKit/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/All-The-Vibes%2FATV-StarterKit/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":32323215,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-26T23:26:28.701Z","status":"online","status_checked_at":"2026-04-27T02:00:06.769Z","response_time":128,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agentic-coding","developer-tools","github-copilot","starter-kit"],"created_at":"2026-04-06T21:01:53.861Z","updated_at":"2026-04-27T05:01:24.158Z","avatar_url":"https://github.com/All-The-Vibes.png","language":"Go","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cp align=\"center\"\u003e\r\n       \u003cimg src=\"./assets/hero-retro.svg\" alt=\"ATV — All The Vibes 2.0 Starter Kit\" width=\"100%\" /\u003e\r\n\u003c/p\u003e\r\n\r\n\u003ch1 align=\"center\"\u003eATV — All The Vibes 2.0 Starter Kit\u003c/h1\u003e\r\n\r\n\u003cp align=\"center\"\u003e\u003cstrong\u003eOne command. Full agentic coding setup. Maximum tasteful chaos.\u003c/strong\u003e\u003c/p\u003e\r\n\r\n\u003cp align=\"center\"\u003e\r\n       \u003ca href=\"https://blazingbeard.github.io/quests/atv-starterkit.html\"\u003e\u003cimg src=\"https://img.shields.io/badge/🎮_New%3F_Start_the_Guided_Training_Quest-ff8c00?style=for-the-badge\" alt=\"Start ATV Quest\"\u003e\u003c/a\u003e\r\n\u003c/p\u003e\r\n\r\n\u003cp align=\"center\"\u003e\r\n       \u003ca href=\"#quick-start\"\u003eQuick start\u003c/a\u003e ·\r\n       \u003ca href=\"#installation\"\u003eInstallation\u003c/a\u003e ·\r\n       \u003ca href=\"docs/marketplace.md\"\u003eMarketplace\u003c/a\u003e ·\r\n       \u003ca href=\"#uninstalling\"\u003eUninstalling\u003c/a\u003e ·\r\n       \u003ca href=\"#the-three-pillars\"\u003eThree pillars\u003c/a\u003e ·\r\n       \u003ca href=\"#the-full-sprint\"\u003eFull sprint\u003c/a\u003e ·\r\n       \u003ca href=\"#how-learning-works\"\u003eLearning\u003c/a\u003e ·\r\n       \u003ca href=\"https://blazingbeard.github.io/quests/atv-starterkit.html\"\u003e🎮 Training Quest\u003c/a\u003e ·\r\n       \u003ca href=\"#development\"\u003eDevelopment\u003c/a\u003e\r\n\u003c/p\u003e\r\n\r\n\u003cvideo src=\"https://github.com/user-attachments/assets/7b6bf18a-2bab-482b-a72d-fac9ab7436c2\" width=\"100%\" autoplay loop muted playsinline controls\u003e\u003c/video\u003e\r\n\r\n---\r\n\r\n## What is ATV 2.0?\r\n\r\nATV 2.0 is a one-command installer that wires together three open-source systems into a single coherent agentic coding environment for GitHub Copilot — grounded in the behavioral principles from [Andrej Karpathy's observations](https://x.com/karpathy/status/2015883857489522876) on LLM coding pitfalls:\r\n\r\n- **[Compound Engineering](https://github.com/EveryInc/compound-engineering-plugin)** — planning-to-knowledge pipeline\r\n- **[gstack](https://github.com/garrytan/gstack)** — sprint execution engine (by Garry Tan / Y Combinator)\r\n- **[agent-browser](https://github.com/vercel-labs/agent-browser)** — browser automation layer (by Vercel)\r\n- **[Karpathy Guidelines](https://github.com/forrestchang/andrej-karpathy-skills)** — behavioral guardrails: think before coding, simplicity first, surgical changes, goal-driven execution\r\n\r\nTogether they cover the full software lifecycle — from \"what should I build?\" through \"is it healthy in production?\" — with 45+ skills, 29 agents, and a learning system that makes your repo smarter with every session.\r\n\r\n---\r\n\r\n## Quick Start\r\n\r\n**Project install** (scaffolds files into your repo, team-shared):\r\n\r\n```bash\r\ncd your-project\r\nnpx atv-starterkit@latest init           # auto-detect stack, install everything\r\nnpx atv-starterkit@latest init --guided  # interactive TUI with multi-stack selection\r\nnpx atv-starterkit@latest uninstall      # cleanly remove everything ATV installed\r\n```\r\n\r\n**Personal install** (Copilot CLI plugin marketplace, follows you across projects):\r\n\r\n```bash\r\ncopilot plugin marketplace add All-The-Vibes/ATV-StarterKit\r\ncopilot plugin install atv-everything@atv-starter-kit\r\n```\r\n\r\nBoth paths can coexist. See [Installation](#installation) for the decision matrix and [docs/marketplace.md](docs/marketplace.md) for category bundles and per-skill plugins.\r\n\r\nThen open **Copilot Chat** (⌃⌘I / Ctrl+Shift+I) and go:\r\n\r\n```text\r\n/ce-brainstorm   →  Explore the problem, produce a design doc\r\n/ce-plan         →  Generate an implementation plan with acceptance criteria\r\n/ce-work         →  Build against the plan with incremental commits\r\n/ce-review       →  Multi-agent code review (security, architecture, performance)\r\n/ce-compound     →  Document what you learned for future sessions\r\n\r\n/lfg             →  Run the full pipeline in one shot\r\n\r\n/atv-doctor      →  Diagnose ATV install health\r\n/atv-update      →  Update ATV marketplace plugins\r\n```\r\n\r\n---\r\n\r\n## Installation\r\n\r\nATV ships in **two flavours** — pick whichever matches your need (or use both):\r\n\r\n| | `npx atv-starterkit init` | Copilot CLI marketplace |\r\n|---|---|---|\r\n| **Files land in** | Your project's `.github/`, `.vscode/`, `docs/` | `~/.copilot/installed-plugins/` |\r\n| **Scope** | Project-level, committed, team-shared | Personal, follows you across projects |\r\n| **What ships** | Skills + agents + MCP + hooks + instructions + setup-steps + docs | Skills + agents only |\r\n| **Best for** | Bootstrapping a new repo, codifying team workflow | Personal cross-project skills |\r\n\r\n### Path 1 — npm scaffold (project-level, recommended for teams)\r\n\r\n```bash\r\nnpx atv-starterkit@latest init       # quick run — downloads binary automatically\r\nnpm install -g atv-starterkit        # or global install\r\natv-starterkit init                  # then run from anywhere\r\n```\r\n\r\nThe npm package downloads the correct platform binary from [GitHub Releases](https://github.com/All-The-Vibes/ATV-StarterKit/releases) — no Go toolchain needed.\r\n\r\n#### Binary (direct download)\r\n\r\nGrab a pre-built binary from [GitHub Releases](https://github.com/All-The-Vibes/ATV-StarterKit/releases/latest) for your platform (macOS, Linux, Windows — amd64/arm64).\r\n\r\n#### From source\r\n\r\n```bash\r\ngit clone https://github.com/All-The-Vibes/ATV-StarterKit.git\r\ncd ATV-StarterKit \u0026\u0026 go build -o atv-installer .\r\n```\r\n\r\n### Path 2 — Copilot CLI marketplace (personal, cross-project)\r\n\r\n```bash\r\ncopilot plugin marketplace add All-The-Vibes/ATV-StarterKit\r\ncopilot plugin install atv-everything@atv-starter-kit       # all skills + all agents\r\n```\r\n\r\nOr pick a category bundle / single skill — full tier breakdown in **[docs/marketplace.md](docs/marketplace.md)**:\r\n\r\n```bash\r\ncopilot plugin install atv-pack-planning@atv-starter-kit    # one category\r\ncopilot plugin install atv-skill-autoresearch@atv-starter-kit  # one skill\r\n```\r\n\r\nThe marketplace ships skills + agents only. For MCP config, hooks, instructions templates, and docs scaffolding use Path 1.\r\n\r\n### Prerequisites\r\n\r\n**Required:** Git, Node.js 16+ (for Path 1).\r\n\r\n**Optional:**\r\n- **Bun** — for gstack browser skills (`/gstack-qa`, `/gstack-browse`, `/gstack-benchmark`)\r\n- **GitHub PAT** — for GitHub MCP server\r\n- **Azure CLI** — for Azure MCP server\r\n- **Copilot CLI** — for Path 2 (`copilot` command)\r\n\r\nWithout Bun, text-based gstack skills still work. `agent-browser` works independently of Bun.\r\n\r\n### Uninstalling\r\n\r\n```bash\r\nnpx atv-starterkit@latest uninstall          # remove ATV files, preserve user-modified configs\r\nnpx atv-starterkit@latest uninstall --force  # remove everything including modified files\r\n```\r\n\r\nRemoves `.github/skills/`, `.github/agents/`, `.github/hooks/`, `.github/copilot-*` config files, `.gstack/`, `.atv/`, and empty doc directories. Files you've customized since installation are preserved by default (checksum comparison against the install manifest). `.vscode/` is never touched.\r\n\r\n---\r\n\r\n## The Three Pillars\r\n\r\n### Karpathy Guidelines — the behavioral foundation\r\n\r\nEvery skill and agent in ATV operates under four principles derived from [Andrej Karpathy's observations](https://x.com/karpathy/status/2015883857489522876) on how LLMs fail at coding. These are installed as a skill (`.github/skills/karpathy-guidelines/SKILL.md`) and shape how Copilot approaches all work:\r\n\r\n| Principle | What it prevents |\r\n|---|---|\r\n| **Think Before Coding** | Wrong assumptions, hidden confusion, silently picking one interpretation |\r\n| **Simplicity First** | Overcomplication, bloated abstractions, speculative features |\r\n| **Surgical Changes** | Drive-by refactoring, touching code you shouldn't, cosmetic \"improvements\" |\r\n| **Goal-Driven Execution** | Vague success criteria, no verification loop, \"make it work\" without checking |\r\n\r\nThese aren't just instructions — they're the operating contract between you and the AI. Without them, Copilot tends toward the exact pitfalls Karpathy described: \"The models make wrong assumptions on your behalf and just run along with them.\"\r\n\r\n### Autoresearch — autonomous experimentation loop\r\n\r\nFor tasks with a measurable metric — performance tuning, test pass rate, bundle size, latency, build time — `/autoresearch` runs an autonomous loop: define the goal + metric + scope, the agent works on a dedicated `autoresearch/\u003ctag\u003e` branch, committing each experiment, running the metric command, and keeping or reverting based on the result. Every experiment is logged to `results.tsv` so you can audit the research trail when the loop ends.\r\n\r\nInstalled as a skill (`.github/skills/autoresearch/SKILL.md`). Sourced from [github/awesome-copilot](https://github.com/github/awesome-copilot/blob/main/skills/autoresearch/SKILL.md) (MIT) by [@luiscantero](https://github.com/luiscantero), inspired by [Karpathy's autoresearch](https://github.com/karpathy/autoresearch).\r\n\r\n**Use when** you have a measurable outcome and want the agent to hill-climb autonomously. **Skip for** one-shot tasks, simple bug fixes, or anything without a clear metric.\r\n\r\n### Compound Engineering — knowledge compounds\r\n\r\nA gated pipeline where each step produces an artifact the next step consumes:\r\n\r\n```text\r\n/ce-brainstorm → /ce-plan → /ce-work → /ce-review → /ce-compound\r\n```\r\n\r\nEvery time you run `/ce-compound`, solved problems get saved to `docs/solutions/`. Next time `/ce-plan` runs, the `learnings-researcher` agent searches those files first. Your repo gets smarter with every PR.\r\n\r\n### gstack — the AI sprint process\r\n\r\n30 slash-command skills covering office hours, engineering review, browser QA, shipping, deploy verification, security audits, and weekly retros. gstack doesn't just give the AI more tools — it gives it a *role*. `/gstack-review` acts as a staff engineer. `/gstack-cso` acts as a chief security officer. The skills are opinionated engineering processes encoded as markdown.\r\n\r\nIncludes safety guardrails (`/gstack-careful`, `/gstack-freeze`, `/gstack-guard`) that prevent destructive commands like `rm -rf` or force-pushes.\r\n\r\n### agent-browser — the eyes of the agent\r\n\r\nA native Rust CLI that controls Chrome via CDP with ~100ms latency. Uses snapshot refs (`@e1`, `@e2`) for deterministic element selection — no CSS selectors or XPath needed. The `open → snapshot → interact → re-snapshot` workflow fits cleanly into an LLM's tool-calling loop.\r\n\r\n---\r\n\r\n## The Guided Experience\r\n\r\nThe guided installer (`--guided`) walks you through four screens:\r\n\r\n**1. Stack Packs** — Multi-select your stacks (TypeScript, Python, Rails). Auto-detected packs are pre-selected.\r\n\r\n**2. Preset** — Choose your depth:\r\n\r\n| Preset | What you get |\r\n|---|---|\r\n| **Starter** | Core CE workflow (13 skills). No network calls, instant install. |\r\n| **Pro** | + gstack sprint skills (35+ skills total) |\r\n| **Full** | + browser QA, benchmarks, agent-browser, Chrome (45+ skills). Requires Bun. |\r\n\r\n**3. Customize** — Power users can drill into category-grouped multi-select. Beginners skip straight to install.\r\n\r\n**4. Install + Summary** — Real-time progress with structured telemetry, then actionable next steps.\r\n\r\n```text\r\n  ✅ Scaffolding ATV files (24 files created, 8 directories) · 340ms\r\n  ⚠️  Syncing gstack skills — fell back to markdown-only · 2.1s\r\n  ✅ Installing agent-browser (CLI ready, skill copied) · 1.8s\r\n\r\n  🎉 ATV Starter Kit ready!\r\n  Install state saved to .atv/install-manifest.json\r\n```\r\n\r\n---\r\n\r\n## The Full Sprint\r\n\r\nEvery skill maps to a phase of the development lifecycle:\r\n\r\n\u003ctable\u003e\r\n       \u003ctr\u003e\r\n              \u003ctd width=\"25%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e💭 Think\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003eFrame the problem\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/ce-brainstorm\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-office-hours\u003c/code\u003e\r\n              \u003c/td\u003e\r\n              \u003ctd width=\"25%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e📋 Plan\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003ePressure-test the approach\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/ce-plan\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-plan-ceo-review\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-plan-eng-review\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-plan-design-review\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-autoplan\u003c/code\u003e\r\n              \u003c/td\u003e\r\n              \u003ctd width=\"25%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e🔨 Build\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003eExecute with momentum\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/ce-work\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/lfg\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/slfg\u003c/code\u003e\r\n              \u003c/td\u003e\r\n              \u003ctd width=\"25%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e👀 Review\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003eFind what you missed\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/ce-review\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-review\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-cso\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-codex\u003c/code\u003e\r\n              \u003c/td\u003e\r\n       \u003c/tr\u003e\r\n       \u003ctr\u003e\r\n              \u003ctd width=\"33.33%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e🧪 Test\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003eUse real browser eyes\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003eagent-browser\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-qa\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-benchmark\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-browse\u003c/code\u003e\r\n              \u003c/td\u003e\r\n              \u003ctd width=\"33.33%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e🚀 Ship\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003eLand without chaos\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-ship\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-land-and-deploy\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-canary\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-document-release\u003c/code\u003e\r\n              \u003c/td\u003e\r\n              \u003ctd width=\"33.33%\" valign=\"top\"\u003e\r\n                     \u003cstrong\u003e📊 Reflect\u003c/strong\u003e\u003cbr /\u003e\r\n                     \u003csub\u003eCompound what you learned\u003c/sub\u003e\u003cbr /\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/ce-compound\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/learn\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/evolve\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/unslop\u003c/code\u003e\u003cbr /\u003e\r\n                     \u003ccode\u003e/gstack-retro\u003c/code\u003e\r\n              \u003c/td\u003e\r\n       \u003c/tr\u003e\r\n\u003c/table\u003e\r\n\r\n### `/lfg` — full pipeline, one command\r\n\r\nEach step must produce output before the next starts (plan file exists, plan was deepened, code was changed). Retries on failure.\r\n\r\n```\r\nplan → deepen → work → review → unslop → resolve → test → video → compound\r\n  ✓       ✓       ✓\r\n```\r\n\r\n### `/slfg` — parallel swarm variant\r\n\r\nSame steps. Planning is sequential, review + test + unslop run in parallel.\r\n\r\n```\r\nplan → deepen → work (swarm) ──→ review    ⎤              resolve → unslop fix → video → compound\r\n                                  test     ⎥ (parallel) →\r\n                                  unslop   ⎦\r\n```\r\n\r\n`unslop fix` removes AI slop after review. `compound` saves learnings for future `ce-plan` runs.\r\n\r\n\u003cdetails\u003e\r\n\u003csummary\u003e\u003cstrong\u003eFull skill reference (45 skills)\u003c/strong\u003e\u003c/summary\u003e\r\n\r\n### Think\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/ce-brainstorm` | Interactive dialogue to clarify requirements; produces design docs in `docs/brainstorms/` |\r\n| `/gstack-office-hours` | YC-style forcing questions that challenge your framing before you write code |\r\n| `/gstack-plan-ceo-review` | CEO-level review: find the 10-star product hiding in the request |\r\n\r\n### Plan\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/ce-plan` | Parallel research agents scan codebase + external docs; auto-discovers brainstorms; outputs plans with acceptance criteria |\r\n| `/deepen-plan` | Enriches each plan section with best practices and performance guidance |\r\n| `/gstack-plan-eng-review` | Forces hidden assumptions into the open: architecture, data flow, edge cases |\r\n| `/gstack-plan-design-review` | Scores design quality 0-10 per dimension; rewrites plan to hit 10 |\r\n| `/gstack-autoplan` | Runs CEO → design → eng review in one command |\r\n\r\n### Build\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/ce-work` | Implements against the plan with incremental commits and system-wide sanity checks |\r\n| `/lfg` | Full pipeline: plan → deepen → work → review → test → video → compound |\r\n| `/slfg` | Parallelized version via swarm agents |\r\n\r\n### Review\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/ce-review` | Parallel review agents: security, performance, architecture, language-specific |\r\n| `/gstack-review` | Staff-level code review with auto-fix and completeness checks |\r\n| `/gstack-design-review` | Design audit with atomic fix commits |\r\n| `/gstack-cso` | OWASP Top 10 + STRIDE threat model |\r\n| `/gstack-codex` | Cross-model review via OpenAI Codex CLI |\r\n\r\n### Test\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `agent-browser` | Direct browser automation: open, snapshot, click, fill, screenshot, inspect |\r\n| `/gstack-qa` | Full QA loop: find bugs in real browser, fix them, write regressions, re-verify |\r\n| `/gstack-qa-only` | Report-only QA (no fixes) |\r\n| `/gstack-benchmark` | Page load baselines, Core Web Vitals, resource sizes |\r\n| `/gstack-browse` | Persistent browser runtime for deeper sessions |\r\n\r\n### Ship\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/gstack-ship` | Sync main, run tests, audit coverage, push, open PR |\r\n| `/gstack-land-and-deploy` | Merge → CI → deploy → verify production |\r\n| `/gstack-canary` | Post-deploy monitoring for errors and regressions |\r\n| `/gstack-document-release` | Auto-update project docs to match what shipped |\r\n\r\n### Reflect\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/ce-compound` | Documents solved problems in `docs/solutions/` for future sessions |\r\n| `/learn` | Extracts coding patterns from recent work into instincts with confidence scoring |\r\n| `/instincts` | Dashboard showing all learned patterns grouped by domain |\r\n| `/evolve` | Promotes mature instincts (confidence \u003e0.8) into permanent Copilot skills |\r\n| `/observe` | Focused pattern analysis on a specific domain or file pattern |\r\n| `/unslop` | De-slop pass: code simplification + comment rot + design slop detection |\r\n| `/gstack-retro` | Team-aware weekly retro with per-person breakdowns |\r\n| `/gstack-learn` | Per-project self-learning infrastructure |\r\n\r\n### Safety Guardrails\r\n\r\n| Skill | What it does |\r\n|---|---|\r\n| `/gstack-careful` | Warns before `rm -rf`, `DROP TABLE`, force-push |\r\n| `/gstack-freeze` | Restricts edits to one directory while debugging |\r\n| `/gstack-guard` | Careful + Freeze combined |\r\n| `/gstack-investigate` | No fixes without systematic investigation first |\r\n\r\n\u003c/details\u003e\r\n\r\n---\r\n\r\n## How Learning Works\r\n\r\nMost AI coding tools treat every session as day one. ATV remembers.\r\n\r\nEvery time you start a Copilot session, the AI has no memory of how *your team* writes code — that you wrap errors with `%w`, prefer table-driven tests, or use constructor injection. ATV fixes this with a **continuous learning pipeline** that observes how you code, extracts reusable patterns, and graduates proven ones into permanent Copilot skills.\r\n\r\n### The Loop\r\n\r\n```text\r\nYou code normally\r\n     ↓\r\nObserver hooks silently capture tool use → .atv/observations.jsonl\r\n     ↓\r\n/learn analyzes observations + git history → instincts with confidence scores\r\n     ↓\r\nConfidence grows with each session (0.5 → 0.6 → 0.7 → 0.8)\r\n     ↓\r\n/evolve promotes mature instincts → .github/skills/learned-*/SKILL.md\r\n     ↓\r\nNext session: Copilot already knows your patterns\r\n```\r\n\r\n### Observer Hooks\r\n\r\nATV installs hooks for all 6 Copilot lifecycle events (`sessionStart`, `sessionEnd`, `preToolUse`, `postToolUse`, `userPromptSubmitted`, `errorOccurred`). A lightweight Node.js script captures every tool interaction to `.atv/observations.jsonl` — silently, with zero impact on your workflow.\r\n\r\n### Instincts\r\n\r\n`/learn` analyzes git history, diffs, observations, and existing solutions to find recurring patterns. Each becomes an \"instinct\" with a confidence score:\r\n\r\n```yaml\r\n# .atv/instincts/project.yaml\r\ninstincts:\r\n  - id: always-wrap-errors\r\n    trigger: \"when returning an error from a function\"\r\n    behavior: \"wrap with fmt.Errorf using %w\"\r\n    confidence: 0.85\r\n    observations: 12\r\n```\r\n\r\nRun `/instincts` to see the dashboard:\r\n\r\n```text\r\n  Error Handling (2 instincts)\r\n    ★ always-wrap-errors        0.9  \"wrap errors with fmt.Errorf %w\"    15 obs\r\n    ● sentinel-errors           0.6  \"use sentinel errors for expected\"   5 obs\r\n\r\n  Testing (1 instinct)\r\n    ★ table-driven-tests        0.85 \"use table-driven test pattern\"     12 obs\r\n\r\n  Legend: ★ ready to evolve (\u003e0.8)  ● active  ○ tentative (\u003c0.5)\r\n```\r\n\r\nWhen an instinct reaches \u003e0.8 confidence, `/evolve` promotes it into a full SKILL.md at `.github/skills/learned-*/`. Copilot auto-discovers these — your AI assistant now *permanently knows* your team's conventions.\r\n\r\n### Design Decisions\r\n\r\n- **Instincts are committed to git** — the whole team benefits, not just one developer\r\n- **Observations are gitignored** — raw data is ephemeral, instincts are permanent\r\n- **Generated skills use `learned-` prefix** — visually distinct from hand-written skills\r\n- **Confidence scoring prevents noise** — only well-established patterns get promoted\r\n\r\n---\r\n\r\n## De-Slop\r\n\r\nAI coding assistants have a tell: over-abstraction, `// This function handles the logic for...` comments, purple-to-blue gradients. Code review catches bugs — but nobody catches *slop*.\r\n\r\n`/unslop` runs three parallel analysis passes on your recent changes:\r\n\r\n```text\r\n/unslop                          →  Report slop in changed files\r\n/unslop src/components/          →  Scope to a directory\r\n/unslop fix                      →  Auto-apply safe fixes\r\n```\r\n\r\n| Pass | What it catches | Example |\r\n|------|----------------|---------|\r\n| **Code Slop** | Over-abstraction, YAGNI violations, nested ternaries | Interface used once → inline it |\r\n| **Comment Rot** | Obvious restatements, AI filler phrases, stale TODOs | `// This function handles auth` → delete |\r\n| **Design Slop** | Generic gradients, template layouts, missing hover states | Purple-to-blue default → use brand palette |\r\n\r\n`/unslop` is wired into both autonomous pipelines — `/lfg` runs `/unslop fix` after review, and `/slfg` runs the report pass in parallel with `ce-review` and browser testing for zero added wall-clock time.\r\n\r\n`/ce-review` asks \"is this correct?\" — `/unslop` asks \"does this look human-written?\" Run both.\r\n\r\n---\r\n\r\n## Memory Architecture\r\n\r\nATV builds seven layers of memory across three reinforcing cycles:\r\n\r\n| Layer | Where | Timescale |\r\n|---|---|---|\r\n| **Observations** | `.atv/observations.jsonl` | Per-session (gitignored) |\r\n| **Instincts** | `.atv/instincts/project.yaml` | Grows every session |\r\n| **Evolved skills** | `.github/skills/learned-*/` | Permanent |\r\n| **Institutional knowledge** | `docs/solutions/*.md` | Permanent |\r\n| **Design decisions** | `docs/brainstorms/*.md` | Permanent |\r\n| **Implementation plans** | `docs/plans/*.md` | Per-feature |\r\n| **Install manifest** | `.atv/install-manifest.json` | Per-install |\r\n\r\n**How they reinforce each other:**\r\n\r\n- **Knowledge compounding** (per-PR): `/ce-compound` saves solved problems → future `/ce-plan` finds them via `learnings-researcher` → fewer repeated mistakes\r\n- **Pattern learning** (per-session): observer hooks → `/learn` → instincts → `/evolve` → permanent skills → Copilot knows your conventions\r\n- **Team propagation** (per-commit): instincts are committed to git → the whole team inherits learned patterns without a style guide\r\n\r\nOver weeks, your repo develops a memory that makes every Copilot session more effective than the last.\r\n\r\n---\r\n\r\n## Agents\r\n\r\n29 specialized agents in `.github/agents/`, invoked by skills during review, planning, learning, and debugging:\r\n\r\n| Category | Agents |\r\n|---|---|\r\n| **Code Review** | `kieran-rails-reviewer`, `kieran-python-reviewer`, `kieran-typescript-reviewer`, `dhh-rails-reviewer`, `code-simplicity-reviewer`, `julik-frontend-races-reviewer` |\r\n| **Security** | `security-sentinel` |\r\n| **Architecture** | `architecture-strategist` |\r\n| **Performance** | `performance-oracle` |\r\n| **Data** | `data-integrity-guardian`, `data-migration-expert`, `schema-drift-detector`, `deployment-verification-agent` |\r\n| **Design** | `design-implementation-reviewer`, `design-iterator`, `figma-design-sync` |\r\n| **Research** | `repo-research-analyst`, `best-practices-researcher`, `framework-docs-researcher`, `learnings-researcher`, `git-history-analyzer` |\r\n| **Process** | `pr-comment-resolver`, `spec-flow-analyzer`, `bug-reproduction-validator`, `pattern-recognition-specialist` |\r\n| **Learning** | `pattern-observer` |\r\n| **Meta** | `agent-native-reviewer`, `ankane-readme-writer` |\r\n| **Ops** | `lint` |\r\n\r\n---\r\n\r\n## What Gets Installed\r\n\r\n### Copilot Integration Points\r\n\r\n| File | Purpose |\r\n|---|---|\r\n| `.github/copilot-instructions.md` | System instructions loaded into every chat |\r\n| `.github/copilot-setup-steps.yml` | Coding Agent initialization steps |\r\n| `.github/copilot-mcp-config.json` | MCP server configuration |\r\n| `.github/skills/*/SKILL.md` | Skills auto-discovered by description match |\r\n| `.github/agents/*.agent.md` | Agents for subagent orchestration |\r\n| `.github/*.instructions.md` | File-scoped instructions via `applyTo` globs |\r\n| `.github/hooks/copilot-hooks.json` | Observer hooks (silent, every tool use) |\r\n\r\n### Supported Stacks\r\n\r\n| Stack | Detection | Additions |\r\n|---|---|---|\r\n| **TypeScript** | `tsconfig.json` | TypeScript reviewer, TS file instructions |\r\n| **Python** | `pyproject.toml` / `requirements.txt` | Python reviewer, Python file instructions |\r\n| **Rails** | `Gemfile` + `config/routes.rb` | 8 Rails-specific agents, Ruby file instructions |\r\n| **General** | fallback | Universal agents and skills |\r\n\r\n### MCP Servers\r\n\r\n| Server | Type | Package |\r\n|---|---|---|\r\n| **Context7** | SSE | `mcp.context7.com` |\r\n| **GitHub** | stdio | `@modelcontextprotocol/server-github` |\r\n| **Azure** | stdio | `@azure/mcp` |\r\n| **Terraform** | stdio | `terraform-mcp-server` |\r\n\r\n---\r\n\r\n## How It Works Under the Hood\r\n\r\n```text\r\natv-installer init --guided\r\n        │\r\n        ▼\r\n Detect stack + prerequisites (git, bun, node)\r\n        │\r\n        ▼\r\n Stack Packs → Preset → Customize?\r\n        │\r\n        ▼\r\n Install with structured telemetry:\r\n        │\r\n        ├── ATV scaffold ──► Embedded templates → .github/skills/*/SKILL.md\r\n        │\r\n        ├── Learning pipeline ──► Observer hooks + skills + instinct storage\r\n        │\r\n        ├── gstack ──► git clone → .gstack/ (staging, gitignored)\r\n        │               └── Copy SKILL.md → .github/skills/gstack-*/\r\n        │\r\n        └── agent-browser ──► npm install -g → agent-browser install (Chrome)\r\n                              └── .github/skills/agent-browser/SKILL.md\r\n        │\r\n        ▼\r\n Write manifest to .atv/install-manifest.json\r\n```\r\n\r\nAll templates are embedded at compile time — no runtime network calls for the core scaffold. gstack requires a network clone (~22MB). Re-running is idempotent: existing files are skipped, JSON configs are merged.\r\n\r\n---\r\n\r\n## Development\r\n\r\n```bash\r\ngo build -o atv-installer .             # build\r\ngo test ./...                            # all tests\r\ngo test ./pkg/installstate/ -v           # manifest + recommendations tests\r\ngo test ./pkg/monitor/ -v                # watcher + drift detection tests\r\ngo test ./test/sandbox/ -v               # integration tests (E2E scenarios)\r\n```\r\n\r\n## Limitations\r\n\r\n- **Bun required for browser skills** — `/gstack-qa`, `/gstack-browse`, `/gstack-benchmark`\r\n- **Network required for gstack** — clones ~22MB at install time\r\n- **gstack setup on Windows** — falls back to `bun run gen:skill-docs` (bash path issues)\r\n- **Token-heavy pipelines** — long multi-agent sessions can hit context limits\r\n\r\n---\r\n\r\n\u003cdiv align=\"center\"\u003e\r\n\r\nMIT — Built by [All The Vibes](https://github.com/All-The-Vibes)\r\n\r\nPowered by [Compound Engineering](https://github.com/EveryInc/compound-engineering-plugin) · [gstack](https://github.com/garrytan/gstack) · [agent-browser](https://github.com/vercel-labs/agent-browser) · [Karpathy Guidelines](https://github.com/forrestchang/andrej-karpathy-skills)\r\n\r\nSpecial thanks to [blazingbeard](https://github.com/blazingbeard) for building out the [guided training quest](https://blazingbeard.github.io/quests/atv-starterkit.html).\r\n\r\n\u003c/div\u003e\r\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fall-the-vibes%2Fatv-starterkit","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fall-the-vibes%2Fatv-starterkit","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fall-the-vibes%2Fatv-starterkit/lists"}