deploy-prototype

ProtoLab hosts throwaway static HTML prototypes. A deploy is a single authenticated HTTP call — no repo, no wrangler, no Cloudflare account on this side. Everything goes through the bundled client:

python3 <skill-dir>/scripts/protolab.py <verb> ...

Prefer the script over hand-rolled curl: it handles config resolution, zipping, pairing, and turns API errors into actionable messages.

A deploy is user-visible and overwrites by slug, so it is never run without explicit user approval of the deploy plan. See "Confirm before deploying" below — this gate applies to every deploy, including re-deploys.

Scope guard — static HTML only

ProtoLab serves static files. If the prototype needs a backend, database, auth, server-side rendering, or a build step (React with JSX, bundlers, npm build), this is the wrong tool. Say so plainly and stop — do not try to "make it work" by building locally and deploying the output unless the user explicitly asks for that.

Which lab? (resolution order)

The script resolves the target lab in this order, first match wins:

1. PROTOLAB_URL + PROTOLAB_TOKEN env vars — explicit override, wins outright
2. PROTOLAB_LAB=<name> env var selecting a config entry
3. --lab <name> — pass this when the user names a lab ("deploy to the work lab")
4. default from config
5. Exactly one lab configured — use it
6. Otherwise the script exits with code 3 and lists the labs. Ask the user which lab they mean; never guess. Config lives at ~/.config/protolab/config (TOML, chmod 600). If that is missing, the script falls back to ./.protolab in the working directory — useful in sandboxed sessions where $HOME does not persist. When pairing in such a session, pass --local to write ./.protolab instead, and make sure .protolab is gitignored.

Verbs

First run — pairing

If the user has never set up (status shows no labs, or deploy exits with code 3 and an empty lab list):

1. Ask for the lab URL only — nothing else, noting to the user that ProtoLab needs to be setup separately first if they have not already done so (https://github.com/birdsigh/ProtoLab).
2. Run pair --url <url>. It prints a short code and blocks, polling.
3. Relay to the user verbatim: "Open <url>/settings and approve code XXXX." The code matters — approval is by code, not hostname.
4. On approval the script saves the token. Ask the user what to call this lab if --name wasn't given (the script suggests the domain's short name); the first lab paired becomes the default. Re-pairing an already-configured URL updates that entry in place — the script dedupes by URL, so this is also the fix for a revoked token.

Confirm before deploying — mandatory

Never run deploy until the user has explicitly approved the deploy plan for this specific deploy. This is a hard gate, not a courtesy: a deploy is publicly reachable at a URL and replace semantics mean a colliding slug silently overwrites whatever was there.

Before every deploy, resolve the target (lab, source, slug) and present the plan back to the user, then wait for approval. Surface all of:

• Lab — the resolved lab's alias and its host URL, so it's unambiguous which lab receives this (e.g. work-lab → https://lab.example.com).
• Source — the exact folder or file being deployed (the path).
• Result URL — the slug and the full URL it produces (e.g. → https://lab.example.com/my-idea/).
• Overwrite — whether that slug already exists on the lab. If it does, say plainly that deploying replaces the existing prototype at that URL. If you can't tell without a network call, say the slug may already be in use and that deploy overwrites on collision. What counts as approval:
• An explicit go-ahead after the plan is shown ("yes", "ship it", "go ahead") — this is the normal path.
• A same-turn instruction that itself fully specifies and authorises the deploy ("deploy ./foo to work-lab as bar, go ahead") can satisfy the gate, but only when lab, source, and slug are all unambiguous. Still echo the resolved plan in your reply so the user sees the URL and overwrite status before it happens. What does NOT count: silence, a vague earlier "let's deploy this at some point", or approval you inferred rather than the user gave. If any of lab, source, or slug is ambiguous, resolve it (ask, per the resolution rules above) before presenting the plan — never guess your way past the gate.

After the deploy, echo the full URL the script returns (see below).

Deploying

• A folder must contain index.html at its root. The script zips it (skipping .DS_Store and friends) and PUTs it. A single .html file is sent directly and becomes index.html.
• If --slug is omitted the script derives one from the folder/file name (lowercased, invalid characters became hyphens). The derived slug is part of the deploy plan the user approves (above), so it's always surfaced before deploy — pay extra attention when it looks surprising, since slugs are the URL and a colliding slug overwrites.
• Limits: 25 MB zipped, 500 files. Slugs: lowercase alphanumeric plus hyphens, max 63 chars; settings, api, favicon.ico, robots.txt, and _-prefixes are reserved.
• Always echo the full URL the script prints (e.g. https://lab.example.com/my-idea/), never just the slug — slugs are per-lab namespaces and mean nothing without the host.

Errors worth translating

• Exit code 4 (HTTP 401): the token was revoked or the lab was reset. Tell the user "the token for '<alias>' was revoked — want me to re-pair?" rather than surfacing the raw error. Re-pair with the same URL.
• Exit code 3: ambiguous lab — ask, never guess.
• rate limited during pairing: the lab throttles pairing hard; wait a minute before retrying.
• Zip/size/slug validation errors are printed with the reason; fix and retry rather than working around the limits. remove-lab only edits local config — it does not revoke the token. If the user wants the token dead, point them at the lab's /settings page.