verify-in-smoke
Skill1 starsChanged 48 days agoMITView on GitHub
Verify a `pnpm smoke` run, screenshot, or dev-server observation before treating it as proof of correctness — checks for a stale sandbox, a squatted port from a sibling worktree, leaked browser state, the sandbox `HOME`, and full-suite flakiness. Use before declaring a screenshot, first-run behavior, or a `pnpm test` result verified, or when `pnpm smoke` behaves unexpectedly (a connect gate that never shows, a refused port, a failing fetch, stale state after a restart, unrelated test failures).
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- worktreeSkill
- find-duplicate-abstractionsSkill
- verify-uiSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 40
Stop the stack with pnpm smoke:stop.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Verify in smoke
Five things make a pnpm smoke run, screenshot, or test result lie to you. Check the one that matches what you're looking at before calling it evidence.
Wrong environment: pnpm dev instead of pnpm smoke
pnpm dev reads the real ~/.maestro, which usually already holds config — first-run behavior (the connect gate) never renders there. → Use pnpm smoke for anything first-run; it wipes .maestro-sandbox on start (ADR-0010).
Wrong address: the port belongs to the worktree
Each worktree serves on its own pair of ports, derived from its path — there is no fixed 5173. A URL copied from a doc, another session, or an open tab points at whatever worktree owns that port. → Ask this worktree: pnpm cockpit:url, or read the line pnpm dev/pnpm smoke prints on start. Screenshot only that address, against a cockpit you started here.
Under pnpm smoke, vite listens on [::1] only. → Open the localhost URL pnpm cockpit:url prints; 127.0.0.1:<web port> is refused while the stack is up.
Owning it at start does not settle it. → Run pnpm smoke:check before each screenshot; it re-asks who holds this worktree's two ports — including the one the browser renders from — and seeds nothing, so it is safe to repeat. It also measures every table screen at several panel widths, with a detail pane open, and names each table wider than its scroll container.
Wrong session: leaked browser state
localStorage is scoped to the origin, so a worktree keeps its own — but a value left by an earlier run of this worktree survives a restart and reads as a product bug. → Clear it before trusting: agent-browser --session <name> storage local clear, then agent-browser --session <name> reload.
The default agent-browser session is shared by every worktree; another session can move its tab. → Drive this worktree's own session, agent-browser --session maestro-<worktree folder>. pnpm cockpit:snapshot "<selector>" [path] reads that session as trimmed text and prints its name and URL.
Wrong home: the sandbox HOME
--smoke redirects HOME, so git runs without a credential helper. A private https fetch then fails could not read Username, while the same fetch works by hand. → Point the sandbox clone at a local bare mirror with url.<mirror>.insteadOf <origin url>, as scripts/fixture-harness.mjs does for the fixture harness. Judge the fetch path only from a run that reaches its origin.
The redirected HOME also gives corepack a cold cache: a hang, or getaddrinfo ENOTFOUND registry.npmjs.org. → A script that redirects HOME carries COREPACK_HOME over and sets COREPACK_ENABLE_DOWNLOAD_PROMPT=0, as scripts/dev.mjs does.
Wrong signal: chained commands flake
pnpm test run standalone is honest; chained behind lint/typecheck in one invocation has shown failures in files the diff never touched. → When pnpm test fails in untouched files, re-run it alone before calling it a regression. Don't chain lint/typecheck/test in one invocation when the result is about to become a green/red claim.
When done
Stop the stack with pnpm smoke:stop. A running stack slows every other worktree's test run.
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
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.

