Build a project with an agent, from idea to ship
For a vibe coder whose project is growing faster than they can keep in their head: plan before code, keep context between sessions, test in a real browser, review before you ship.
- 1
Clarify, write a spec and a plan
Makes the agent ask questions and write a spec before it codes; the plan lives in files, so it survives a full context.
- 2
Up-to-date docs for every library
Pulls the docs of the version you use into the agent, so it stops inventing APIs.
Needs: A free API key raises the limits
- 3
Find your way in a large codebase
Code search in plain words that returns only the parts that matter, so the agent spends fewer tokens reading.
- 4
Remember the project between sessions
Writes decisions and findings into linked Markdown notes you can read and edit yourself.
Alternatives
- 5
Keep the code small
Steers the agent away from over-engineering: less code to read, review and pay for.
Alternatives
- 6
A UI that does not look AI-made
Art direction, references and a visual check in the browser. design-dna measures how far your colours drift from your brand.
- 7
Test in a real browser, on a phone too
The agent opens your pages, clicks through them and reads what is on screen. Airsc audits its own pages with it.
Recommended Airsc used it
- 8
Review for security before you ship
Skills from Trail of Bits, a security audit firm, for finding the bugs that matter.
Alternatives
- 9
Ship and keep an eye on it
Issues, pull requests and CI from the agent; uptime and a status page once it is live.
Alternatives
- 10
See what the agent costs
Reads your session files locally and splits the spend by project, model and task.
Alternatives
Picks reviewed by hand: a recommended resource is rated Safe with a quality score of 85 or more and updated in the last 90 days. Alternatives marked Caution are worth reading about first.