Agent-first
Runbooks your coding agent follows from setup to live payments
SaaScaffold ships with 40 runbooks in agents/. Tell your agent what you want: AGENTS.md sends it to the right runbook, which says what to check, which commands to run and where to stop and ask you.
Say what you want
The agent maps a plain request to a runbook and keeps to that runbook’s scope.
| You say | The agent starts from | How far it goes |
|---|---|---|
| Prepare the environment | lifecycle/01-project-setup.md | Prepares only. Nothing is released. |
| Check whether we can launch | lifecycle/04-pre-launch-checklist.md | Checks only, and says what it would change before fixing anything. |
| Deploy it, take it live | lifecycle/05-production-release.md | Releases the checked version. A first launch keeps payments off or in admin test mode. |
| Configure it after the deploy | lifecycle/06-after-launch.md | Configures and checks each module, with no real charges. |
| What’s left? | site/checklist.md | Decides which items apply to your site, then works through them stage by stage. |
| Open live payments | lifecycle/07-payment-live.md | Only when you ask, with the products, amounts and providers named. |
| Set up a module | integrations/, operations/ | That module only. |
| Write or review a public page | seo-geo/on-page-seo.md | Only the pages involved, citing the rule it applied. |
Tell your agent what you want. The runbooks tell it how.
Your coding agent reads the runbooks, edits the code, runs the commands and records what it did. It stops and asks you before it deploys, and it never handles your secret keys.
The agent reads AGENTS.md, puts your name and copy into the site config, and runs the tests.
Seven stages from setup to live payments
Each stage has its own runbook and a clear finish line.
- 01Environment setupResources, identities, configuration and gaps are listed. Anything not yet released is called out.
- 02DevelopmentCore tasks, pages, layouts, permissions and their checks have evidence.
- 03SEO and GEOThe page list, public HTML, languages and discovery are checked and fixed.
- 04Pre-launchPre-deploy items pass, live items wait for the deploy, and what stays switched off is clear.
- 05ReleaseA named version goes out through the controlled release. The live site is checked separately.
- 06After launchEach module is checked for real, and the payment mode follows the first-launch rules.
- 07Live paymentsReal transactions, credits and the final switch are checked within the authorisation you gave.
Where it stops and asks you
Reading a runbook never authorises the risky steps inside it.
- Every deploy needs your go-ahead. The agent first names the target site and the full commit.
- New production SQL is shown to you for review before it runs.
- It never prints or stores secret keys. It tells you the variable name and where to set it.
- It opens live payments only when you ask, and never makes a real charge on its own.
- It does not reset or delete your local data without your consent.
- It leaves alone the servers and databases you started, and stops the ones it started.
It leaves a record you can read
What the agent learns about your site goes into four files in site/, committed with the code and free of secrets. The next agent, or you, picks up from there.
| File | What it answers |
|---|---|
site/profile.md | What the site is: owner details and the non-secret IDs of outside accounts. |
site/checklist.md | How far the build has got, item by item, with links to the records. |
site/deployment.md | What is actually running in each environment: version, migrations, resources, which secrets are set, switches and payment mode. |
site/runs/YYYY-MM-DD-<task>.md | What one run did: authorisation, target, step results, changes and blockers. Added, never rewritten. |
# After launch: Stripe test payments connected
- Time: 2026-11-18 09:30–10:05 (UTC)
- Agent: Claude Code
- Authorisation: "Set up Stripe in test mode"; not included: live payments
- Target: production, https://pixelforge.example, Worker pixel-forge, D1
- Source: a1b2c3d…, clean working tree
- Runbook: agents/integrations/payment-setup.md
## Results
| Step | Result | Notes |
| ----------------------------- | -------- | ---------------------------------- |
| Product credits-500 mapped | Passed | Stripe price ID saved in admin |
| Test connection | Passed | secretKey, webhookSecret: set |
| Admin test purchase | Passed | Order ord_8f2…, 500 credits |
| Live purchases | Not done | Outside this authorisation |
## Written back
- site/checklist.md: A9 → saved
- site/deployment.md: payment mode admin_testpnpm release writes its own record and updates the deployment file, whether it finishes or stops. Screenshots and raw logs stay out of Git.
Works with the agent you already use
Claude Code, Codex or any agent that reads AGENTS.md. The runbooks are plain Markdown in agents/, so a person can follow them too.
Get the SaaScaffold source code
Buy once, then claim access to the private repository with your GitHub username.
BuySee how extending works