Files

T

Marcos Paulo 48c31dc1b3 build: migrate from npm to pnpm

Ten git worktrees each carried their own 225 MB node_modules (1.1 GB across
five) and paid 11s per `npm ci`. pnpm hardlinks from a shared store: the same
five worktrees cost ~250 MB total, and a fresh install is 4s.

What changed beyond the mechanical rename:

- `overrides` moved to `pnpm-workspace.yaml`. pnpm 11 does not read the `pnpm`
  field in package.json *or* npm's top-level `overrides`, and it fails silently
  — the vite/defu/language-server pins would have quietly stopped applying.
- Build scripts are blocked by default in pnpm; esbuild and sharp are allowed
  explicitly via `allowBuilds` (renamed from `onlyBuiltDependencies` in 11).
- `packageManager` + `engines` pin the toolchain.
- gate.sh rejects a package-lock.json/yarn.lock/bun.lock outright, so an agent
  running `npm install` out of habit fails loudly instead of building a second,
  divergent dependency tree.
- CI bootstraps pnpm with `npm install --global pnpm@11.25.0` rather than
  corepack (unbundled as of Node 25) or pnpm/action-setup (this self-hosted
  act-runner has never run a job; fetching a third-party action is not
  something to discover on the first one).

Two pre-existing CI bugs fixed while in the file:

- the gate installed with `npm install --package-lock=false`, which discarded
  the lockfile the previous session had just fixed.
- the visual-regression step imported `playwright`, which is not a dependency,
  and `visual-regression.mjs` has no compare mode anyway — in CI it overwrote
  its own baselines and passed unconditionally. Removed with a comment; it
  comes back when it can diff.

The `publish` job is now manual (`workflow_dispatch`). During the migration
dist/ holds three HTML files against the live pages branch's ten, so publishing
on every push to main would take the site down to a stub. Restore at task 20.

HANDOVER.md's incident log still says npm where it describes what happened at
the time; that is history, not a missed rename.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

2026-09-05 04:29:42 +00:00

3.0 KiB

Raw Blame History

Rule: quality gates

Three tiers. Each is scoped so that many agents committing in parallel worktrees stay fast — the whole point is that a gate you are tempted to skip is not a gate.

Tier	Hook	Scope	Budget	Runs
1	`pre-commit`	staged files only (lint-staged)	< 3s	every commit
2	`pre-push`	whole project: build + verify + audit	< 90s	every push
3	CI (Gitea Actions)	tier 2 + visual regression	minutes	every push to `main`

Tier 1 must stay under a few seconds. If it creeps, move the check to tier 2. An agent that waits 40s per commit will start passing --no-verify, and then you have no gate at all.

Tier 1 — pre-commit (lint-staged)

Formats and lints only what you staged. Scoped by construction, so ten parallel worktrees do ten small jobs, not ten full-project sweeps.

Prettier + ESLint on *.{js,mjs,ts,astro}
Prettier + Stylelint on *.css
check-tokens.mjs on changed .astro/.css — catches raw hex before it lands
Prettier on *.{json,md}

Tier 2 — pre-push

The real gate:

astro check        types
astro build        it compiles
verify.mjs         42 content assertions — count must not fall
audit-ui.mjs       no external runtime dependencies
check-tokens.mjs   full sweep

Tier 3 — CI

Tier 2 plus screenshot comparison against .agents/snapshots/. Only CI has the budget for it.

Bypassing

--no-verify is allowed exactly once: a work-in-progress commit on your own task branch that you will rebase away. It is never allowed on a commit you intend to merge, and the pre-push gate has no bypass.

If a gate is wrong, fix the gate in its own commit. Do not route around it.

Parallelism

Hooks are per-worktree. Git's index.lock is per-worktree, so parallel commits do not contend.
The heavy tier-2 gate takes a lock (.git/af-gate.lock, shared across worktrees) so ten agents pushing at once do not run ten concurrent builds and thrash the machine. Waiters queue; they do not fail.
pnpm install --frozen-lockfile in a fresh worktree is cheap: pnpm hardlinks from the shared content-addressable store, so a second worktree costs seconds and almost no disk instead of another 225 MB. No --prefer-offline needed.

The silent-failure mode you must know about

Husky sets core.hooksPath to .husky/_, and .husky/_ is generated by pnpm install, not committed. A fresh git worktree add therefore has hooks configured but the directory missing — so hooks silently do not run. Every commit passes. Nothing is checked.

.agents/scripts/worktree.sh start runs the install and then verifies. Check manually any time you did not use it:

.agents/scripts/verify-hooks.sh

Run this first in any worktree you did not create with the script.

3.0 KiB Raw Blame History