chore: add fallow config and fix high-signal findings

[jrusso1020]

What

Add a .fallowrc.jsonc and fix the genuine code-quality issues fallow surfaces. Drops dead-code findings from 601 → 276 (45% noise reduction from config, the rest are real signal that's either fixed here or listed as deferred follow-ups).

Why

npx fallow --summary on main was reporting 601 dead-code issues, but ~half were false positives: docs .mdx files flagged as unused, vitest/bun-test files not registered as entry points, worker entry points loaded via file path invisible to static analysis, deps consumed dynamically (tsup external, peer/static-file in tests, readFileSync reads of @fontsource/*). Without config, fallow can't be a useful signal; with config, the remaining findings are real.

How

Config (.fallowrc.jsonc)

• Declare test/runtime/worker/script entry points that fallow's plugin heuristics miss
• Ignore docs, test fixtures, skill test-corpora, registry, examples, CI fixtures
• Ignore deps that are dynamically referenced (puppeteer-core, esbuild, giget, gsap, happy-dom, ffmpeg-static, ffprobe-static, all @fontsource/*, workspace siblings)
• ignoreExports for intentional namespace barrels (per-namespace ML manager exports; CLI per-command examples convention from CLAUDE.md; isPathInside in fileServer.ts which is test-only and has different symlink semantics from the canonical one in utils/paths.ts)

Inline build:fonts blob extraction
packages/{cli,producer}/package.json had node -e "..." blobs containing {...} that fallow mis-parsed as glob alternate groups, emitting 4 warnings and skipping the real entry. Moved each into a dedicated scripts/build-fonts.mjs.

Duplicate-export fixes

• FileIcon in studio/SystemIcons.tsx — removed the dead alias; FileTreeIcons.tsx has the real, used one
• ValidationResult — gsapParser.ts had an identical-shape duplicate of core.types.ts. Dropped the duplicate; both parsers now import from core.types

Circular dep
manualEditsDom.ts re-exported three clearStudio* functions from manualEditsSnapshot.ts, which imports four helpers from manualEditsDom.ts — a back-edge cycle. Moved the re-export to manualEdits.ts (the package-public barrel) where the rest of the snapshot re-exports already live.

Unused-dep cleanup

• studio: drop motion (no imports anywhere) and codemirror (umbrella package; @codemirror/* sub-packages are used directly)
• cli: drop mime-types (zero imports), its now-orphan src/utils/mime.ts (hardcoded mime table that didn't even use the package), and its tsup external entry

Test plan

• bun run typecheck clean across core, cli, producer, studio
• bunx oxlint + bunx oxfmt --check clean on changed files
• bunx vitest run packages/studio/src/components/editor/manualEdits.test.ts — 18/18 pass (verifies the cycle break doesn't break behavior)
• bunx vitest run packages/core/src/parsers — 69/69 pass (gsapParser + htmlParser tests verify the ValidationResult consolidation)
• bunx vitest run packages/studio — 505/505 pass
• bunx vitest run packages/cli — 332/332 pass
• node scripts/build-fonts.mjs in both packages/cli and packages/producer runs successfully

Deferred follow-ups

Real findings that need their own PRs:

• 8 circular deps in producer/services/render/stages/ — renderOrchestrator ↔ captureHdr* / captureStage / extractVideosStage form a hub cycle. Real refactor.
• ~14 unused files in producer/src/services/ (audioExtractor, audioMixer, browserManager, chunkEncoder, compilationRunner, parallelCoordinator, screenshotService, streamingEncoder, videoFrameExtractor) — most are thin re-export shims to @hyperframes/engine that aren't in the public exports map. Need to confirm no deep-import consumers before deletion.
• waveform.ts complexity hotspot — fallow's start with... suggestion. Separate audit.

[jrusso1020]

• chore: add fallow config and fix high-signal findings #938 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[miguel-heygen]

Clean, well-scoped PR. Every change is justified and the config is well-documented.

Verified:

• FileIcon in SystemIcons.tsx is dead — FileTreeNodes.tsx imports from FileTreeIcons.tsx, not SystemIcons
• mime-types has zero imports across cli/src
• codemirror umbrella and motion have zero imports in studio/src — sub-packages (@codemirror/*) are used directly
• ValidationResult consolidation: core.types.ts is the canonical location, both parsers now import from there
• Circular dep fix: moving the clearStudio* re-exports from manualEditsDom.ts to the manualEdits.ts barrel breaks the back-edge without changing the public API
• build-fonts.mjs extraction: eliminates the fallow glob-parsing false positives from inline node -e blobs

The deferred follow-ups (8 circular deps in producer/services/render/stages, ~14 unused shim files, waveform complexity) are the right call for separate PRs.

CI green except one regression shard still in progress — no code changes that would affect render output.

[vanceingalls]

Solid, well-scoped cleanup. Config is explained (not bulk-allowlisted), each ignore has a one-line rationale, and the structural fixes (cycle break, duplicate-type merge, dead-code drops) are individually verifiable. Public-API surfaces preserved. Recommend ship.

Calibrated strengths

• .fallowrc.jsonc ignore rationales are concrete and load-bearing — e.g. isPathInside carries the "different symlink semantics from utils/paths.ts" note, so the next reader knows the duplication is intentional rather than a latent dedup target. This is exactly the right discipline for a tool config — it's documentation for the next person who hits a fallow flag.
• manualEdits.ts:42-50 — moving the clearStudio* re-export off the back-edge in manualEditsDom.ts:464 and onto the package-public barrel is the structurally correct cycle break. Verified the only external consumer (hooks/useDomEditCommits.ts:11-15) goes through the ./manualEdits barrel, so the public surface is unchanged. manualEdits.test.ts also imports via the barrel — tests cover the new path.
• ValidationResult consolidation: the duplicate at gsapParser.ts:347 is dropped, both parsers now import from core.types, and core/index.ts:20 continues to re-export the canonical one. No deep-import consumers (from "@hyperframes/core/parsers/gsapParser") were depending on the type — verified across the workspace.
• The node -e → scripts/build-fonts.mjs extraction is the right call. Even setting aside fallow's glob-parser issue, the inline blob was unreadable and unreviewable; a dedicated 14-line script with existsSync + execSync is much easier to maintain. Symmetric move in both cli and producer.

Findings

nit — bun.lock carries a stray aws-lambda version bump (0.0.1 → 0.6.22). Not in the PR description, and packages/aws-lambda/package.json doesn't appear in the diff. Looks like an unrelated lockfile resync from running bun install against a workspace whose version fields had drifted on main. Not load-bearing (aws-lambda isn't published as 0.6.x afaict), but worth a sentence in the description so the next person reviewing the lockfile delta isn't confused. Same goes for the synchronized 0.6.15 → 0.6.22 bumps across every workspace package.

nit — mime-db regressed 1.54.0 → 1.52.0 in the lockfile after dropping the direct mime-types dep. The remaining aws-cdk-lib/mime-types chain now wins resolution at the older version. No functional impact (only consumed transitively by aws-cdk-lib), but if some downstream tool reads top-level mime-db it'll see the older version. Worth a glance, but not a blocker.

nit — fallow worker-entry declarations point at file paths that don't exist on main yet (packages/producer/src/services/pngDecodeBlitWorker.ts, shaderTransitionWorker.ts). Verified the typecheck CI is green on this PR, so either (a) those files exist on the branch I haven't traversed, or (b) fallow tolerates missing entries gracefully. Just flagging that adding entries-that-don't-yet-exist sets up a foot-gun if a future rename silently drops them off the entry list — a comment near those two lines noting the companion *Pool.ts files would prevent future drift.

nit — deferred follow-up bullet for the 8 producer/services circular deps is the right move, but consider opening a tracking issue before this merges. "Real refactor, listed in PR body" tends to fall off when the PR merges and the body scrolls off; an issue keeps it indexed.

CI (verbatim, fresh)

{"failed":[],"pending":["regression-shards (shard-1, ...)","regression-shards (shard-2, ...)","regression-shards (shard-3, ...)","regression-shards (shard-4, ...)","regression-shards (shard-5, ...)","regression-shards (shard-6, ...)","regression-shards (shard-7, ...)","regression-shards (shard-8, ...)"]}

All required & non-shard checks green (Build, Typecheck, Lint, Test, CLI smoke, runtime contract, Format, Preflight, file size, semantic title, perf gates, Windows render, preview-regression, CodeQL). 8 regression shards still in progress; nothing failed.

Verdict

Verdict: APPROVE
Reasoning: Every non-trivial deletion verifies — FileIcon alias isn't consumed outside its file (the FileTree.tsx FileIcon is a separate local symbol, not the same export), mime-types / motion / codemirror umbrella have zero imports, and the ValidationResult + cycle-break refactors preserve the public surface and pass typecheck. Config rationales are the right level of detail. The lockfile drift is a noise item, not a correctness one.

Review by Vai

[vanceingalls]

Re-stamp converting the prior COMMENT to a formal APPROVE — content of the review body still applies.

Verified deletions

Every removed symbol grep-verified across workspace including registry/, examples/, docs/:

• FileIcon — only out-of-file ref is a separate local symbol in FileTree.tsx. Clean.
• mime-types, motion, codemirror packages — zero imports across the workspace. Clean.
• MIME_TYPES in cli/utils/mime.ts — file-local only; other files have their own MIME tables. Clean.

Cycle break

useDomEditCommits.ts:11 (only external consumer of clearStudio*) goes through the ./manualEdits barrel, which the PR updates to re-export directly from manualEditsSnapshot. Public surface preserved — no adopter break.

Findings

nit: lockfile churn — every workspace's 0.6.15 → 0.6.22 bump plus a stray aws-lambda 0.0.1 → 0.6.22 bump aren't in the PR body. Either intentional (version sync) or accidental. Worth a line in the description either way.

nit: the fallow tool itself is a nice addition for the team's static-analysis bar. If you intend to keep using it, a bunx fallow pre-commit lefthook or CI step would lock in the cleanup discipline.

Verdict: APPROVE.

Re-stamp by Vai