agentleFS
kyle-chalmerskyle-chalmers/streamsnow/skills/build-app/SKILL.md

build-app

Skill17 starsChanged 32 days agoMITView on GitHub

The front door for building a Streamlit-in-Snowflake app from idea to opened PR, or resuming one mid-build. Orchestrates spec, data discovery, page design, scaffold, a parallel page build and review by subagents, and ship, with human checkpoints between them. Start here for any new app, to document an existing one, to add a page, or to act on feedback about a live app. Use when the user says "build an app", "new dashboard", "add a page", "spec this out", "pick up where we left off", "here's feedback on <app>", "the numbers look wrong" or "polish this dashboard".

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Follow spec.md for apps/<slug>/REQUIREMENTS.md.

    line 37
  2. Distinctive:

    Follow design.md: data-scout profiles the data into §3, app-designer plans every page (forms, copy, glossary, shared data) into §4–§8.

    line 42
  3. Very specific:

    Follow scaffold.md: streamsnow new <domain> <function> (runtime from §9, _shared/runtime-decision.md), then the shared layer from the …

    line 47
  4. Distinctive:

    Follow pages.md § Parallel build: one page-builder per §4 page, dispatched in one message.

    line 53
  5. Distinctive:

    Follow verify.md: preview, walkthrough, then perf-reviewer, viz-critic,

    line 60

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/build-app

Repo overlay: if .streamsnow/overlays/build-app.md exists in this repo, read it first — committed, repo-specific additions/overrides (_shared/overlays.md). Outside Claude Code, also read _shared/other-agents.md.

Lifecycle: spec → discover → design → scaffold → build → preview → verify → ship → done. It reads apps/<slug>/REQUIREMENTS.md §11 to resume and never skips a checkpoint.

Orchestrator. You run every deterministic step (CLI, merges, generation, validation) and own the shared files; subagents get fixed briefs (briefs/) naming their inputs, the only files they may write, and the result they return. The CLI and gates decide pass/fail, never a subagent. No subagents? Follow each brief yourself (_shared/other-agents.md).

Modes

  • Default — a new app, or a slug to resume from §11 Current phase; never restart a build.
  • --spec [<slug>] — write or refresh the spec only, then stop. Covers new specs, tickets and backfill from an existing app's source (automatic when apps/<slug>/ has code). spec.md.
  • <slug> --feedback "<feedback>" — classify feedback on a live app, then fix it via feedback.md.

Phase 0 · Preflight

  1. Report which streamsnow runs and its version, then run streamsnow doctor --format json.
  2. streamsnow not on PATH, a required check failing, or no streamsnow.config.yaml: say "This machine or repo isn't set up yet, so I'm running onboarding first", follow /onboard in full, then continue (--spec may carry on without config). optional failures get one line.

Phase 1 — Spec, then CHECKPOINT 1

  1. Follow spec.md for apps/<slug>/REQUIREMENTS.md. CP1: confirm its one-screen summary (pages and their questions, sources, TTLs, runtime). Block until confirmed.

Phase 2 — Discover and design, then CHECKPOINT 1b

  1. Follow design.md: data-scout profiles the data into §3, app-designer plans every page (forms, copy, glossary, shared data) into §4–§8. CP1b: the text wireframe.

Phase 3 — Scaffold the foundation

  1. Follow scaffold.md: streamsnow new <domain> <function> (runtime from §9, _shared/runtime-decision.md), then the shared layer from the design: glossary, pages/_data.py loaders, their queries, About page; commit on build/<slug>.

Phase 4 — Build pages in parallel

  1. Follow pages.md § Parallel build: one page-builder per §4 page, dispatched in one message; each writes only its page and its own queries and returns its nav, index and glossary entries. Merge them, run streamsnow sql-review generate <slug> once, and commit the round: pages and their review SQL land together.

Phase 5 — Preview, verify, then CHECKPOINT 2

  1. Follow verify.md: preview, walkthrough, then perf-reviewer, viz-critic, streamlit-reviewer, cold-reader in parallel; fixes go to file owners, two rounds max. CP2: the user clicks through every page. Block until they answer.

Phase 6 — Check and ship, then CHECKPOINT 3

  1. streamsnow validate-app <slug> until PASS (/validate-app explains each FAIL); then follow /review-app's instructions in full and present its verdict.
  2. CP3: validation passes, review is clean → ask the user to type /ship-app <slug>; the agent cannot start it. A first deploy may need admin DDL (deploy-setup --admin): surface, don't run.

State — §11 Build Progress

A Current phase line (the lifecycle above) plus an append-only Sessions log whose last line names the next command. Update it on every phase change; never rewrite past lines. On resume, jump to Current phase. done or in-production (backfilled) means the app is live: new §4 pages re-enter at design; feedback uses --feedback, review depth /review-app. Apps started before 0.8 name the old start-app skill in that log; say /build-app instead.

Out of scope

Porting an external app → /migrate-app; review depth → /review-app; live numbers → /sql-review; machine, repo and Snowflake setup → /onboard.

Done when

The PR is open, streamsnow validate-app <slug> passes, and §11 reads Current phase: done.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.