Lessons from running the starter on promptquick.ai for three weeks:
- Email: idempotency keys are unique per environment, since staging and
production can share one provider account and their job IDs overlap.
- Background jobs: the scheduler's own run history is cleaned on a
schedule, so it can't creep toward a free database plan's limit.
- Ship a .gitattributes with LF line endings and the common binary
types named, so the generated-file drift checks pass on Windows;
SETUP.md and the README list it.
- Build: a risky slice is marked done but keeps its acceptance criteria
until its independent review is settled; the Now block notes the
pending review.
- Set up: the owner's steps are click by click, in a working order,
name the exact site and environment, and say when a redeploy is needed.
- Set up: AGENTS.md stays under about 100 lines, not counting blocks
tools add themselves; the starter's own text already left little room
under 80.
- READMEs: the staging and production pipeline is tested on a live
site, and the review covers sign-in, payments, database changes and
live data, as the starter does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The root README's Codex badge now says untested, matching the README
and release v1.0.0. The production line says the agent holds no hosted
keys and the owner pushes and deploys; the deny list is a backstop.
Both READMEs say the starter is instructions with no code: the agent
builds the environment guard, the CI checks and the pipeline.
FUNDING.yml keeps only the buy_me_a_coffee line, and GROWTH-PLAN.md
leaves the public .gitignore.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Rename prompt-templates/ to planning-prompts/ with numbered step
folders, so GitHub lists them in workflow order. Folder, prompt file
and skill now share one name per step (2-prd/prd.md is /prd).
- Merge the full and ultra-lean MVP plan prompts into 5-mvp-plan/, as
the /mvp-plan skill already did.
- Rename the /mvp skill to /mvp-scope; the installer removes an earlier
mvp install and tells the owner about the new name.
- Rename agent-configs/claude-code-desktop/ to agent-configs/claude-code/,
since the rules apply to the terminal too.
- Rewrite each prompt guide in one short layout with previous and next
links, drop outdated model and temperature advice, and fix the copied
"final PRD draft" lines and the claimed v0.dev section.
- Point to GitHub's Copy raw content button for copying a prompt.
- Move the AI App Starter details from the root README into the starter
guide, add the skills to the top table, and align the test plan
wording with what the prompt produces.
- Add issue forms, a Discussions link and a short contributing guide.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The skill proposes an idea, waits for a go, shows storyboard frames,
waits again, then builds and checks the video with Remotion. brief.md
holds the craft and copy rules. INSTALL.md and the README now also
explain how to install only some skills.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Replace the starter's three problem tables with five short points.
- Mark Codex as beta, drop the duplicate nav line, trim the overview table.
- Point newcomers to the PRD prompt, and explain the effort setting.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Add starters/nextjs-supabase-vercel: a project an AI coding agent builds
with its owner in five stages (plan, decide, set up, build, launch), with
local-only development, blocked hosted commands, just-in-time questions,
rules-not-recipes stage files and modules for payments, email and jobs.
- Rewrite the root README around the planning prompts, the AI App Starter
and agent configurations.
- Remove the session continuation and v0.dev prompts, and the permission
guardrails config.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Updated final output structure guidelines to emphasize brevity and clarity, specifying a maximum of 6 sentences and the use of Simplified Technical English.
