.cursorrules and contribute.md still teach yarn — missed by the pnpm migration (#2069)

[NieZhuZhu]

Summary

The Yarn→pnpm migration (#2069, merged Jun 12) updated AGENTS.md, CLAUDE.md, app/AGENTS.md and removed the root .yarnrc.yml — but two files that also carry command instructions were missed and still teach yarn:

.cursorrules (auto-loaded by Cursor as agent instructions), in "Build & Deployment → Scripts" / "Dependencies":

• L147: `yarn ios` / `yarn android` for platform-specific builds
• L148: `yarn test` for Jest testing
• L149: `yarn test:e2e:ios` / `yarn test:e2e:android` for E2E
• L153: Yarn workspaces for monorepo management

(zero mentions of pnpm anywhere in the file's 190 lines)

contribute.md — L37 yarn format, L40 yarn lint, L43 yarn workspaces foreach …, and the PR checklist items L50–51 ("yarn lint passes" / "yarn format has been run").

Why it matters

These aren't incidental prose — they're the exact commands a Cursor/AI agent (or a new contributor following contribute.md) will actually run, and ground truth says they can't work as documented:

• root package.json: "packageManager": "pnpm@11.7.0+…" (plus a lint:pnpm-version script enforcing it)
• pnpm-lock.yaml at root; no yarn.lock anywhere in the repo
• app/package.json's own ios / android / test scripts invoke pnpm … internally
• app/AGENTS.md already teaches pnpm ios / pnpm test / … — so .cursorrules contradicts its sibling doc for the very same directory

An agent that loads .cursorrules is told to use a package manager the repo explicitly forbids (root AGENTS.md: "pnpm only … never use npm or yarn").

Suggested fix

• .cursorrules L147–149: yarn → pnpm; L153: "Yarn workspaces" → "pnpm workspaces".
• contribute.md: same substitution on the five lines above (the yarn workspaces foreach invocation would become the pnpm -r / pnpm --filter equivalent).
• Optional sweep while at it: app/.yarnrc.yml and circuits/.yarnrc.yml look like inert pre-migration leftovers (no yarn.lock or packageManager: yarn anywhere references them).

I have the patch ready and am happy to open a PR against dev — filing an issue first because the contribution guide discourages docs-only PRs, and I'd rather let you decide whether agent-facing command docs fall under that policy or under "commands that don't work".

Disclosure

Found while validating ai-harness-doctor, a docs-vs-lockfile drift linter we maintain — mentioning for transparency. Everything above is verifiable directly from the cited files (checked at main HEAD 15b167e).