plate

Imported from udecode/plate (AGENTS.md). Install upstream with npx skills add udecode/plate. Copyright stays with the author.

• .agents/AGENTS.md and .agents/rules/*.mdc are source of truth. After editing them, run pnpm install to sync. Never edit SKILL.md directly.
• In all interactions and commit messages, be extremely concise and sacrifice grammar for the sake of concision.
• Answer in English by default. Switch languages only when the user explicitly asks for another language.
• Prefer the best long-term architecture fix over the nearest local patch. If the real fix is an API or abstraction change, do that.

Git

• Git: Never git add, commit, push, or create PR unless the user explicitly asks, or the active command/skill explicitly requires it.
• Task PR default: The task and major-task skills explicitly require verified code-changing work to be committed, pushed, and opened/updated as a PR unless the user explicitly says not to, the work has no local patch, or a real blocker is recorded. Do not treat the lack of a separate "open a PR" user sentence as a blocker.
• One PR, one task: Every PR requires its own task invocation and dedicated task plan. The PR body must name that plan, the plan must exist at the PR head, and it must identify the exact PR. A batch plan may order PRs, but one batch-level task or autoclosure run never substitutes.
• Noncompliant PR: autoclosure must comment and close any PR without valid per-PR task evidence. The comment must explain the requirement, show how to run task, and recommend GPT-5.6 with high-or-higher reasoning effort. The comment must succeed before closing; read back both the comment and CLOSED state, then stop without reviewing, repairing, or merging the PR.
• Open PR follow-up: If the current branch already has an open PR and you make any change, treat that PR as explicit authorization to commit and push the entire checkout before handoff. Do not leave local-only follow-up changes on an open PR branch.
• Push scope: When you do commit and push, include unrelated dirty files outside src; those are often manual user changes or synced skill/docs updates, so do not silently leave them behind.
• PR: Before creating or updating a PR, run check. If it fails, stop and fix it or report the blocker. Do not open a PR with failing check unless the user explicitly says to.
• PR branch: If the user explicitly says to open or create a PR, do not ask for confirmation. If the current branch is main, create a new codex/ branch first, then commit/push/open the PR. If already on a non-main branch, proceed directly.
• Merge override: If the user explicitly says to merge it, do it. Do not wait for CI to turn green, do not ask again, and use admin merge if that is what it takes.
• Dirty workspace: Never pause to ask about unrelated local changes. Continue work and ignore unrelated diffs.
• Never browse GitHub files. For library/API questions or unfamiliar deps, inspect the repo at ..; if missing, clone https://github.com/{owner}/{repo}.git to ../{repo-name}.

Packages

• DX: Optimize for the absolute best developer experience. JSDoc must be first-class for agents. Every API surface should be intuitive for both humans and AI agents.
• Docs: NEVER write changelog-style language ("has been removed", "new feature", "previously", "now supports"). Docs are user-facing reference for the LATEST state only. Write as if no prior version exists. No migration notes, no "what changed" — just document what IS. Follow .agents/rules/docs-creator.mdc for writing tone/structure.
• Templates: templates/** is CI-controlled output. Never manually edit or commit template source, manifests, or lockfiles. Fix the source registry, package, or workflow inputs and let CI regenerate templates. If local verification rewrites template files, restore them before handoff.
• Barrels: If you change package exports, move public files, add/remove files under exported folders, or CI says pnpm brl produced changes, run pnpm brl before final verification/commit and include the generated barrel updates.
• Do not write TDD cases for dead code/legacy removal assertions (for example: "should not contain old API X anymore"). Remove the dead path directly and keep tests focused on current behavior.
• Prefer inline when used once; extract constants only when reused.

Tooling

• Never run build:registry outside CI. Registry build output is automated by CI and does not belong in local agent commits.
• If typecheck/build/dev suddenly blows up with missing-module or package-resolution garbage that does not match the current diff, run pnpm run reinstall once before deeper debugging.
• Treat local-only React runtime weirdness as install corruption first, not product code:
  • Invalid hook call
  • resolveDispatcher() / null dispatcher crashes
  • package-local node_modules/react or node_modules/react-dom paths under packages/*
  • mixed .bun and .pnpm React paths in the same failing stack
• If pnpm test, bun test, or pnpm check suddenly fails with those signals and the failure does not line up with the current diff, run pnpm run reinstall once before blocking on the task.
• pnpm run reinstall is the repo reset button: it deletes root/workspace/app node_modules, .turbo, apps/www/.next, and tsconfig.tsbuildinfo, then runs pnpm install.
• Do not use pnpm run reinstall as a lazy substitute for fixing real code errors.
• For react-dnd / DnD fixes, do not treat a follow-up Bun Invalid hook call, resolveDispatcher(), or mixed .bun + .pnpm React stack as proof the DnD fix is wrong. In this repo, run pnpm run reinstall once before reopening the diagnosis; that failure shape is usually local env rot, not duplicate deps or broken DnD logic.

Skill

Use those skills when relevant:

• autogoal for any prompt with a verifiable and quantitative outcome. Always use the autogoal skill before durable work when the task has a measurable completion threshold
• orchestrator when the current thread should route per-branch work to child threads instead of executing locally
• task for normal repo task execution. It is mandatory once per PR, including every PR inside a batch
• major-task for heavyweight architecture, framework comparison, migration, benchmark, or proposal work
• autoclosure to finish the current tree without creating new product scope, or to comment and close a PR that lacks verifiable per-PR task evidence
• clawsweeper ftextor Slate issue-ledger triage, duplicate/stale/invalid classification, small high-confidence issue processing, and exact claim sync
• clawpatch for Clawpatch init/map/review/report/fix/revalidate workflows
• editor-test-harvester for mining external editor repositories for portable editor-behavior tests, Slate v2 coverage gaps, and copy/refactor/create decisions
• editor-harvest-plan for turning an editor-test-harvester result into a lane-specific Slate v2 or Plate execution plan
• sync-plate-ui for fork-aware Plate UI registry component syncs into downstream apps like Potion, including status, planning, review, dashboard, and accepted-row apply workflows
• release-lanes for beta/latest release lane maintenance, promote, direct main-to-next sync, beta pre-mode, and npm/GitHub release verification
• sync-main-to-next for the fast direct main -> next release-lane sync wrapper without promotion or autoreview ceremony
• tdd
• resolve-pr-feedback owns actionable GitHub PR feedback after source-backed triage
• For .agents/**, .claude/**, .codex/**, skills, hooks, commands, prompts, or user-action tooling, use the autogoal agent-native pack, run agent-native-reviewer, then end with autoreview
• Repository-owner authorization: task, major-task, and agent-native workflows may invoke their required final autoreview without per-run confirmation
• @.agents/rules/changeset.mdc when updating packages to write a changeset before completing
• @.agents/rules/plate-plan.mdc when defining or updating editor-behavior law, authority maps, protocol rows, or parity coverage

Plate-specific CE exclusions:

• Do not install or reference these by default in this repo unless the user explicitly asks: data-integrity-guardian, data-migration-expert, data-migrations-reviewer, schema-drift-detector, deployment-verification-agent, dhh-rails-reviewer, kieran-rails-reviewer, kieran-python-reviewer, previous-comments-reviewer, pr-comment-resolver, figma-design-sync.
• Reason: Plate is a framework/editor repo. Data migration, Rails, deployment, PR-thread, and Figma workflow agents are mostly overkill or the wrong shape here.

Goal plans:

• For issue-backed goal work, start the filename with the ticket number. Example: docs/plans/DEV-4510-fix-schema.md
• For non-ticket goal work, keep the date-based format. Example: docs/plans/2026-02-07-fix-schema.md

Browser usage:

• When updating content/**, apps/www/**, or packages/**, start the relevant dev server and verify the affected route, UI, or package-facing behavior with [@Browser](plugin://browser@openai-bundled) before handoff. If the surface has no runnable browser path or the server/browser is blocked, say that explicitly.
• Always try [@browser-use](plugin://browser-use@openai-bundled) first for browser usage.
• Do not substitute Puppeteer, standalone Playwright, or raw Chrome DevTools for browser usage.
• For Plate registry/browser proof, prefer /blocks/[id]-demo over docs wrappers when that standalone demo route exists.

Commands

Slate v2 sibling repo

• In .tmp/slate-v2 dir, keep bun check fast: lint, typecheck, and unit/package tests only.
• Do not put bun test:integration-local in bun check; it is a closure/release gate, not an iteration gate.
• Use bun check:full when a local full browser sweep is needed.
• bun check:full must include release-proof guards before the full browser sweep: release discipline, slate-browser proof contracts, scoped mobile proof, persistent-profile soak, then bun test:integration-local.
• Use bun test:mobile-device-proof:raw only on a machine/device lane that can provide real Appium Android/iOS proof artifacts. Do not let semantic mobile handles or Playwright mobile viewport rows satisfy raw-device claims.
• During editor-kernel/browser work, use focused package tests and focused Playwright greps first.
• Run bun test:integration-local only before marking an architecture/browser plan done, before a release-quality browser claim, or when explicitly requested.

Development

Default to source-first typecheck. Do not build packages just to run types unless the repo script or failure proves the typecheck graph still resolves built dist output.

If typecheck fails with stale workspace-package declarations, source/dist split-brain, or unresolved package exports, first inspect the package/app paths and source-entry setup. Build only when the affected surface intentionally validates release artifacts or still has no source-first typecheck path.

If a local-only build/runtime/test failure points at corrupted files under node_modules/.bun, mixed .bun / .pnpm React installs, package-local node_modules/react* symlinks, Invalid hook call, or other non-versioned env state while CI is green, clean local env before changing repo code: run pnpm run reinstall once, then rerun the exact failing command. If the failure shape changes or disappears, it was local env rot. If not, go back to normal debugging.

Required sequence for type checking modified packages:

1. pnpm install - Install dependencies when needed by the task or lockfile state.
2. pnpm turbo typecheck --filter=./packages/modified-package - Run source-first package type checking.
3. If that fails because the graph resolves built output, fix the source-entry or paths setup when that is the right long-term shape.
4. Build only when checking artifact output, package exports, or a package that intentionally has no source-first typecheck path.
5. pnpm lint:fix - Auto-fix linting issues.

For multiple modified packages:

# Typecheck multiple specific packages through their source graph
pnpm turbo typecheck --filter=./packages/core --filter=./packages/utils

# Lint multiple packages
pnpm lint:fix

Alternative approaches:

# Typecheck since last commit
pnpm turbo typecheck --filter='[HEAD^1]'

# Typecheck all changed packages in current branch
pnpm turbo typecheck --filter='...[origin/main]'

# For workspace-specific operations
pnpm --filter @platejs/core typecheck
pnpm --filter @platejs/core lint:fix

Full project commands (use only if needed, these are very slow):

• pnpm build - Build all packages (only use when necessary)
• pnpm typecheck - Root package typecheck. It should use source-first package graphs; if it needs a build, treat that as source-entry debt unless the check is explicitly artifact-facing.
• bun run test - Run the fast default test suite during iteration
• bun test - Run the full test suite only at the end of the complete task