2D platformer in the browser
A side-scrolling jump-and-run in the style of the classics, with art and sound you are allowed to ship.
13 parts · 10 checks in the definition of done · test-built twice, last on Oct 12, 2026
A kit is a tested recipe for one kind of project. Each job in it gets a part from AgentGid's catalogs with the reason it was picked, its exact version and its license, and the kit writes the files your coding agent builds from: AGENTS.md with a plan and a check after every step, a first prompt, CREDITS.md and the MCP setup for Claude Code, Cursor, Codex or Copilot. A recipe is listed only after a coding agent built the project from the kit files alone.
New to kits? Read how to use a kit, from the download to the first prompt.
A side-scrolling jump-and-run in the style of the classics, with art and sound you are allowed to ship.
13 parts · 10 checks in the definition of done · test-built twice, last on Oct 12, 2026
A product, object or small set that visitors can rotate and explore, lit and textured with CC0 assets.
17 parts · 9 checks in the definition of done · test-built twice, last on Oct 12, 2026
A command that turns a script into a vertical video with voiceover, stock footage and captions, and keeps a record of every license.
16 parts · 8 checks in the definition of done · test-built twice, last on Oct 12, 2026
A tool that turns a recording into a clean transcript, subtitle files and, if you want, short notes. The speech model runs on your own machine.
6 parts · 9 checks in the definition of done · test-built twice, last on Oct 12, 2026
Your own PDF toolbox that runs on your machine: files never go to an online converter, and every result is checked against the original.
10 parts · 9 checks in the definition of done · test-built twice, last on Oct 12, 2026
A Telegram bot that follows the RSS feeds you give it and sends a short digest at the time you choose, to you, a group or a channel.
10 parts · 9 checks in the definition of done · test-built twice, last on Oct 12, 2026
A small, tested MCP server that lets Claude Code, Cursor, Codex and other agents search and use your own API or data, with typed tools and safe errors.
10 parts · 7 checks in the definition of done · test-built twice, last on Oct 12, 2026
A Next.js app where people sign up, pay for a plan with Stripe and get email, with the parts every SaaS needs tested before you write your own features.
23 parts · 8 checks in the definition of done · test-built twice, last on Oct 12, 2026
video to text
by @diiiiiima · ♥ 1 · reviewed by AgentGid, not test-built
Short-video pipeline: script to finished file
by @member-d31e52 · reviewed by AgentGid, not test-built
телеграм-бот для записи клиентов в салон красоты с напоминаниями
by @member-d31e52 · reviewed by AgentGid, not test-built
Chrome extension that summarizes the page I am reading with an LLM
by @member-d31e52 · reviewed by AgentGid, not test-built
create a tool for compressing and editing pdf
by @diiiiiima · reviewed by AgentGid, not test-built
A recipe lists the jobs of a project, the candidates for each job, the rules that rule a candidate out, and a build plan with a check after every step. Adding a project type means one recipe file and a test build.
177 of 220 parts come from AgentGid's catalogs of 1,429 tools, 173 asset sources and 422 APIs as they were on Oct 12, 2026. The rest, and 57 packs, models and fonts, were collected from official sources, because the catalogs stop at the level of a source.
Matching your goal, filtering by license and writing the brief are lookups and text assembly. No model is called, so the same request always gives the same kit, and an agent can ask for it through the AgentGid MCP server (plan_project).
A fresh coding agent in a clean sandbox gets only the kit files and builds the project, writing down everything in the brief that was wrong, missing or unclear. Each recipe page shows those findings and what nobody has checked yet.
Describe it in the kit builder. When no tested recipe fits, it can assemble a draft kit: a model splits your goal into jobs and every part still comes from AgentGid's catalogs, marked as not test-built. For single questions, which library, which API or which asset source, use the catalogs: Stack cost calculator, Tools, APIs, Assets.