fix(lint): video and audio take their length from the file, only an img needs data-duration

[miguel-heygen]

What changes

media_missing_duration no longer fires for video and audio. Their length comes from the media file when data-duration is absent, so data-start alone is enough. The rule still reports an img that carries data-start or data-track-index and has no data-duration or data-end: an image has no length of its own, and the runtime gives an unresolved image an open end (packages/core/src/runtime/init.ts around line 833; only video and audio are auto-resolved in packages/core/src/compiler/htmlCompiler.ts). A bare img with no timing attributes stays a static layer and is not flagged. The message now says "image".

timeline_element_missing_timing for non-media elements is unchanged.

Verification

• packages/lint: structure.test.ts and structure.catalog.test.ts pass on a Linux box (15 tests; the catalog test still reports nothing for every registry item and skill example).
• Fixtures: a timed img without a duration is flagged and passes with one; video and audio with data-start and data-track-index and no duration produce no structure finding; a bare img is not flagged.
• Deliberate mutations: removing the video/audio exclusion turns 1 test red, removing the img-only gate turns 2 red, removing the img branch turns 1 red.
• Grepped the repo: the rule id and message are read only by structure.ts and its tests; the Studio lint panel renders findings generically.

Before / After

No visible change: a lint-rule tweak inside packages/lint, with no Studio component, style or markup touched, so there is nothing to capture.

Pre-review

An independent reviewer confirmed the runtime claim at file:line, ran hostile fixtures (uppercase tags, media as a sub-composition host, data-end only, bare, div control) and the three mutations above. No blocking or should-fix findings.

Audited	Trusting	Not exercised
Runtime resolution of img vs video/audio; other readers of the rule id and docs wording; hostile fixtures; 3 mutations; catalog zero-false-positive sweep	Author's Linux-box test runs	Studio's image drop default of 3 s (not opened); CLI/Studio end to end

[somanshreddy]

Reviewed at 2975f712. The deletion is right for two of the three media tags; for img the premise is only half-verified, so one question before I stamp.

What the rule caught (from the removed tests): a media element that IS a timeline clip (has data-start/data-track-index) but lacks data-duration — media_missing_duration — while a static <img> with no timing attrs was already exempt. So this is specifically about timeline-clip media with no length.

Right to remove for video/audio. They resolve to their intrinsic file length at render when data-duration is absent, and the lint is static so it can't know that length anyway — flagging them was a false positive. No concern.

img is where the premise doesn't fully hold, as far as I can confirm. The comment cites a "dropped-image default for img," but the only default I can find — DEFAULT_TIMELINE_ASSET_DURATION — lives in packages/studio/src/utils/studioHelpers.ts, i.e. it's applied when Studio adds a clip (it writes data-duration), not by core at render. I couldn't locate a core/render default that gives a bare, hand-or-agent-authored <img data-start="0"> (no data-duration) a length. If there isn't one, then removing the rule drops the guard for exactly the case it was built for: an agent adds an image clip (start set), forgets the duration, and there's no nudge — the image then renders with whatever core's no-duration fallback is (historically "stays to the end of the composition").

Question before stamp: does core/render give a timeline-clip <img> (data-start, no data-duration) a default length, or is that Studio-add-only?

• If core defaults it → premise holds, delete away, I'll approve.
• If it's Studio-only → consider scoping the removal to video/audio and keeping media_missing_duration for img, since that's the one type with no render default and the one an agent is most likely to under-specify.

(I searched rather than assume, but didn't exhaustively trace core's timing resolver — if you can point at the render-side img default I'll happily flip.)

Codex disabled per workspace cap — manual pass only.

[somanshreddy]

Re-reviewed at 03818e86. The narrowing is exactly right — thanks for scoping it. Holding only for the current CI run to settle (the reds shown are superseded, not real).

Narrowing verified — my earlier finding is fully addressed:

• FILE_LENGTH_TAGS = new Set(["video", "audio"]) — those are excluded from the missing-duration check (their length comes from the file / the compiler probes it), so media_missing_duration no longer fires on video/audio.
• img keeps the rule: isImage = row.tag === "img" still emits media_missing_duration with the sharpened message "an image on the timeline without data-duration, and an image has no length of its own," and the static-layer exemption is now img-specific. So a bare-authored image clip is still nudged toward an explicit duration until the runtime gives it the dropped-image default — which is the sequencing #4144's note calls out.

CI is settling, and the reds aren't real: the Test/Producer/Tests on windows-latest fails are 0-duration — superseded/cancelled by the newer run (the exact misleading-cancelled-red that #4149 is built to hide). The current run's Test/studio-core/Typecheck/Producer: unit tests are pending. For a lint PR the Test job (runs structure.test.ts) is the content-validating check, so I'll flip to approve once it lands green on the current run rather than stamp over a pending run whose 0s predecessors were cancelled.

Code-wise this is an approve. Re-ping me or I'll flip when the current Test is green.

Codex disabled per workspace cap — manual pass only.

[somanshreddy]

Flipping to APPROVE at 03818e86 — the current run's Test is green and the narrowing is exactly right.

• Narrowing (my earlier finding, implemented): FILE_LENGTH_TAGS = {video, audio} are excluded (length from the file / the compiler probes it); img keeps media_missing_duration ("an image has no length of its own") until the runtime gives images the dropped-image default. Correct scoping.
• Test is green on the current run (35431494108, 5m56s full pass — it runs structure.test.ts, the content-validating check for this lint change), plus Test: runtime contract green.
• The two remaining red rows (Tests on windows-latest: matrix, Producer: matrix) are 0-duration matrix-parents from the older, superseded runs (35431494007 / 35431493958) that were cancelled when the current run started — the misleading-cancelled-red #4149 is built to hide. Not real failures at this head.

merge is BLOCKED only on REVIEW_REQUIRED. Approving.

Codex disabled per workspace cap — manual pass only.