Repository navigation
chore(root): sync mmcc/moq with upstream main (10.0.0-beta.32) - #39
Conversation
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
…ges (videojs#2222) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com> Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: Darius Cepulis <dcepulis@mux.com>
Co-authored-by: Claude <noreply@anthropic.com>
Sync with upstream videojs/v10 main (d083950, 10.0.0-beta.32). Resolutions beyond the textual conflicts: - Port the tsdown entries to the Vite+ pack configs: `moq` entry in packages/spf/vite.config.ts; keep simple-moq-video off the CDN via an exclusion in the auto-discovered media entries (html/vite.config.ts). - switchVideoTrack: keep the all-sets candidate pool and confineToActiveSwitchingSet ahead of upstream's stickToSelectedCodecs; merge preferCodecFamilies and playerResolutionCap into the rules. - safeDefine moved to registration/safe-define. - Migrate MoQ tests from 'vitest' to 'vite-plus/test'; run the new jsdoc formatter and padding rule over the MoQ sources. - Move .claude/plans/moq-engine-implementation.md to .agents/plans (.claude/plans is now a generated alias).
📦 Bundle Size Report🎨 @videojs/html
Small changes (42, ≤ 300 B)
Presets (7)
Media (20)
Players (5)
Skins (37)
UI Components (49)
⚛️ @videojs/react
Small changes (37, ≤ 300 B)
Presets (7)
Media (23)
Players (5)
Skins (26)
UI Components (39)
🧩 @videojs/core
Small changes (4, ≤ 300 B)
Entries (76)
🏷️ @videojs/element — 1 small size change
Entries (2)
📦 @videojs/store — no changesEntries (3)
🔧 @videojs/utils
Small changes (4, ≤ 300 B)
Entries (13)
📦 @videojs/media
Small changes (8, ≤ 300 B)
Entries (21)
📦 @videojs/spf
Small changes (4, ≤ 300 B)
Entries (11)
📦 @videojs/jsx
ℹ️ How to interpretEach entry is independently bundled, minified, and brotli-compressed. Initial size includes its static import graph; lazy dynamic chunks are reported separately. Entries are not additive because their dependency graphs overlap. Preset rows represent realistic combined bundles. Changes of 300 B or less across initial, lazy, and total size are collapsed, not discarded. Run |
There was a problem hiding this comment.
💡 Codex Review
This entry exports both createMoqEngine and the MoqMediaElement/MoqMediaMixin facade under @videojs/spf/moq. The scoped repository guide requires media facades to use separate adapter entry points and explicitly prohibits exposing them from engine entries; keeping these surfaces together prevents independent public contracts and bundle budgets. Move the adapter exports to a dedicated media entry and leave this entry engine-only.
AGENTS.md reference: packages/spf/src/AGENTS.md:L16-L16
When a caller configures muted autoplay through the standard IDL path (element.defaultMuted = true), this public field changes only its stored value: it neither adds the muted attribute nor updates element.muted, so playback remains audible and autoplay may be rejected. The declarative attribute path works, but the tests cover only that direction; implement a reflected getter/setter as CustomMediaElement does.
URL schemes are case-insensitive, and parseMoqSource() already accepts an uppercase scheme because new URL() normalizes it, but this guard rejects sources such as MOQT://relay.example/.... In that case setupMoqSession remains in preconditions-unmet and never connects, with no useful error; normalize the scheme or parse the URL before comparing it.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
20 issues found across 3000 files
Not reviewed (too large): site/scripts/api-docs-builder/src/media-element-handler.ts (~2,408 lines) - if these are generated or fixture files, add them to ignored paths to exclude them from future reviews.
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="apps/sandbox/templates/html-mux-background-video/main.ts">
<violation number="1" location="apps/sandbox/templates/html-mux-background-video/main.ts:29">
P2: When the sandbox source picker selects `mp4-1`, this renders its `.mp4` URL into the HLS-only background-video engine, so playback fails instead of showing a valid source. Restrict the background preset's available sources to `type === 'hls'` (while retaining the intentional HLS failure cases such as MPEG-TS and unlicensed DRM).</violation>
</file>
<file name="packages/cli/vite.config.ts">
<violation number="1" location="packages/cli/vite.config.ts:53">
P2: CLI tests use a CDN manifest that does not match the manifest used by the packed CLI, so CDN validation can pass while production accepts or rejects different renderers. Generate the test manifest from the same discovery source, or update the fixture and its expectations whenever the CDN entries change.</violation>
</file>
<file name="apps/sandbox/templates/html-shaka-video/main.ts">
<violation number="1" location="apps/sandbox/templates/html-shaka-video/main.ts:38">
P2: When the iframe starts with a known structured source such as `mux-source-cmaf`, `SOURCES[state.source].url` is undefined and this template sends the literal `"undefined"` to Shaka. Guard the optional URL here, and normalize unsupported sources to a Shaka-compatible default before rendering.</violation>
</file>
<file name="internal/design/spf/use-cases/audio-only-mode-override.md">
<violation number="1" location="internal/design/spf/use-cases/audio-only-mode-override.md:315">
P2: The updated path `media/hls-audio/element.ts` does not exist — `packages/html/src/media/hls-audio/` contains only `index.ts` and `media.ts`, and no `element.ts` exists anywhere under `packages/html`. `HlsAudioElement` is both defined and registered in `define/media/hls-audio.ts`, so the new text also misstates that file's role. Revert to pointing the row at `define/media/hls-audio.ts` so readers can find the export.</violation>
</file>
<file name=".github/workflows/issue-sync.yml">
<violation number="1" location=".github/workflows/issue-sync.yml:37">
P1: When a merged PR contains hidden prompt-injection text, this step gives Codex an issues-write `GH_TOKEN` plus writable workspace and unrestricted network access, so it can close or alter issues outside the PR's intended scope or exfiltrate the token. The warning in the prompt is not an enforcement boundary; run the model without write credentials in a read-only/restricted sandbox and apply only validated structured decisions in a separate step.</violation>
</file>
<file name=".github/workflows/forward-port-changelog.yml">
<violation number="1" location=".github/workflows/forward-port-changelog.yml:20">
P2: When a changelog push overlaps a publishing release, this workflow can push `site/v10` before `cd.yml` performs its final force-push, so the release overwrites the changelog and no later site push retriggers this workflow. Share the `production` concurrency group or otherwise run this job after deployment completes.</violation>
</file>
<file name=".github/workflows/e2e-failure-triage.yml">
<violation number="1" location=".github/workflows/e2e-failure-triage.yml:39">
P1: Because this action processes attacker-controlled logs, artifacts, and pull-request data, enabling workspace writes plus network access lets prompt injection run arbitrary commands, read `GH_TOKEN`, and exfiltrate it. Use a least-privilege permission profile and move required GitHub mutations into fixed workflow steps.</violation>
</file>
<file name="packages/core/scripts/generate-i18n-types.ts">
<violation number="1" location="packages/core/scripts/generate-i18n-types.ts:38">
P3: The log always prints "Updated generated translation types" even when the new `existsSync`/`readFileSync` guard skips the write because the file is already current. In watch/dev runs this reports an update that did not happen. Print only when a write occurred, e.g. move the log inside the condition or log a no-change notice.</violation>
</file>
<file name="internal/design/ui/poster.md">
<violation number="1" location="internal/design/ui/poster.md:31">
P3: 'The one each binding renders' overclaims for the HTML binding. The HTML element renders no image and only sets src on the author's <img> — it cannot attach alt="". Only the React binding renders an img with alt="". Say the default applies to the image the React binding renders, while the HTML binding fills src only (alt stays the author's), consistent with the following sentences about supplying your own alt.</violation>
</file>
<file name=".github/workflows/api-reference-sync.yml">
<violation number="1" location=".github/workflows/api-reference-sync.yml:281">
P3: This job is labeled GPT-5.6 Sol but selects `gpt-5.6-terra`, so it runs a different model than the workflow and PR description advertise. Set the model to `gpt-5.6-sol` or align the label and description with Terra.</violation>
<violation number="2" location=".github/workflows/api-reference-sync.yml:285">
P2: If an untrusted API artifact or repository file contains a prompt injection, this setting lets Codex run arbitrary network commands from a writable workspace while `GH_TOKEN` is available. Run Codex read-only without a write-capable token, then create the issue in a separate trusted step or enforce an equivalent narrow permission profile.</violation>
</file>
<file name=".github/workflows/issue-triage.yml">
<violation number="1" location=".github/workflows/issue-triage.yml:30">
P1: When an external user opens or reopens an issue, `openai/codex-action@v1` rejects the run before Codex because `allow-users` defaults to empty and the actor lacks repository write access. This disables the workflow’s external-issue triage path; add the intended external-user allowlist or route this job through a trusted bot.</violation>
<violation number="2" location=".github/workflows/issue-triage.yml:35">
P1: When a maintainer reopens an attacker-controlled issue, Codex receives its body as prompt data, inherits the write-capable `GH_TOKEN`, and can make arbitrary network requests. A prompt injection can therefore exfiltrate the token or mutate issues; run untrusted issue analysis without credentials or network access, then apply only validated results in a separate trusted step.</violation>
</file>
<file name=".agents/skills/write-docs/references/state-tooling.md">
<violation number="1" location=".agents/skills/write-docs/references/state-tooling.md:280">
P3: The corrected example now uses `player.append(videoElement)` + `document.body.append(player)` and never calls `attach()`, but the section's cause still says "Called method before `attach()`." Update the cause/title to reference the missing media append so the troubleshooting text matches the revised API.</violation>
</file>
<file name=".vite-hooks/commit-msg">
<violation number="1" location=".vite-hooks/commit-msg:1">
P2: The commit-msg hook runs `vp run commitlint`, which executes `commitlint --edit` without forwarding the commit-message file git passes as `$1`. The `--edit` flag without a value falls back to `.git/COMMIT_EDITMSG`, so it validates the wrong file whenever git provides a non-default message path (e.g., worktree/scripted/`--amend` flows). Use conventional hook wiring so the exact message file is validated: pass the path through to commitlint.</violation>
</file>
<file name="apps/sandbox/templates/html-mux-video/main.ts">
<violation number="1" location="apps/sandbox/templates/html-mux-video/main.ts:50">
P2: When `hls-drm` is selected, this leaves the poster on the skin's fallback image, which lacks `crossorigin`, so StackBlitz blocks the cross-origin Mux thumbnail. Render a `crossorigin` slotted image whenever `poster` exists, not only when `placeholder` exists.</violation>
</file>
<file name="apps/sandbox/templates/html-hls-background-video/main.ts">
<violation number="1" location="apps/sandbox/templates/html-hls-background-video/main.ts:25">
P2: Opening this template directly (e.g. `/<template-name>/` per the README) now defaults to source `hls-1`, an MPEG-TS ladder, so the SPF background video shows nothing (SPF has no TS transmuxer). The old page hardcoded a CMAF/fMP4 URL, and the shared `DEFAULT_BACKGROUND_SOURCE = 'hls-4k'` is the CMAF default the shell uses for this preset. Initialize from `DEFAULT_BACKGROUND_SOURCE` (via `?source=`/shell still overrides) so the direct-navigation demo plays instead of failing.</violation>
</file>
<file name="apps/sandbox/package.json">
<violation number="1" location="apps/sandbox/package.json:43">
P2: The StackBlitz `startCommand` runs `pnpm dev`, but this change deleted the package's `dev` script, so the standalone boot (which, per README, uploads only this directory and never the monorepo root) fails with "Missing script: dev". Point the command at a command that exists here, e.g. `vp dev` (matching the `dev` task's `vp dev --host`; the sandboxTemplateSyncPlugin regenerates `src/` on buildStart), or restore a `dev` npm script.</violation>
</file>
<file name=".zed/settings.json">
<violation number="1" location=".zed/settings.json:15">
P2: The Zed oxlint LSP runs with default settings because it has no configPath and the repo has no .oxlintrc.json, while all project lint rules and plugins (anti-slop, typescript/react, rule overrides) live only in vite.config.ts. As a result the in-editor lint/`source.fixAll.oxc` signals will diverge from `pnpm lint`, showing project-disabled rules and omitting the project's custom rules. Previously the biome config enforced `require_config_file` so the editor matched CI; this is a regression. Point the LSP at the project config (e.g. add `"configPath"` for a checked-in oxlint config that mirrors the vite.config.ts lint section) or explicitly scope the oxlint LSP to basic defaults.</violation>
</file>
<file name="packages/core/package.json">
<violation number="1" location="packages/core/package.json:38">
P2: The new "./vjsc" subpath export is only added to the top-level exports, not to publishConfig.exports. npm publishes with the publishConfig.exports object replacing the manifest's exports, so the published @videojs/core will not expose ./vjsc and any consumer importing @videojs/core/vjsc (including packages/skins/vjsc/components/*) will fail to resolve it after publish even though it works in the workspace. Mirror the ./vjsc entry in publishConfig.exports.</violation>
</file>
Note: This PR contains a large number of files. cubic selects up to 200 of the highest-priority eligible files for this review, so some files may not have been reviewed.
Re-trigger cubic
| openai-api-key: ${{ secrets.OPENAI_API_KEY }} | ||
| model: gpt-5.6-terra | ||
| effort: medium | ||
| sandbox: workspace-write |
There was a problem hiding this comment.
P1: When a merged PR contains hidden prompt-injection text, this step gives Codex an issues-write GH_TOKEN plus writable workspace and unrestricted network access, so it can close or alter issues outside the PR's intended scope or exfiltrate the token. The warning in the prompt is not an enforcement boundary; run the model without write credentials in a read-only/restricted sandbox and apply only validated structured decisions in a separate step.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .github/workflows/issue-sync.yml, line 37:
<comment>When a merged PR contains hidden prompt-injection text, this step gives Codex an issues-write `GH_TOKEN` plus writable workspace and unrestricted network access, so it can close or alter issues outside the PR's intended scope or exfiltrate the token. The warning in the prompt is not an enforcement boundary; run the model without write credentials in a read-only/restricted sandbox and apply only validated structured decisions in a separate step.</comment>
<file context>
@@ -19,22 +18,25 @@ jobs:
+ openai-api-key: ${{ secrets.OPENAI_API_KEY }}
+ model: gpt-5.6-terra
+ effort: medium
+ sandbox: workspace-write
+ safety-strategy: drop-sudo
+ codex-args: '["--config", "sandbox_workspace_write.network_access=true"]'
</file context>
| effort: medium | ||
| sandbox: workspace-write | ||
| safety-strategy: drop-sudo | ||
| codex-args: '["--config", "sandbox_workspace_write.network_access=true"]' |
There was a problem hiding this comment.
P1: Because this action processes attacker-controlled logs, artifacts, and pull-request data, enabling workspace writes plus network access lets prompt injection run arbitrary commands, read GH_TOKEN, and exfiltrate it. Use a least-privilege permission profile and move required GitHub mutations into fixed workflow steps.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .github/workflows/e2e-failure-triage.yml, line 39:
<comment>Because this action processes attacker-controlled logs, artifacts, and pull-request data, enabling workspace writes plus network access lets prompt injection run arbitrary commands, read `GH_TOKEN`, and exfiltrate it. Use a least-privilege permission profile and move required GitHub mutations into fixed workflow steps.</comment>
<file context>
@@ -25,17 +26,17 @@ jobs:
+ effort: medium
+ sandbox: workspace-write
+ safety-strategy: drop-sudo
+ codex-args: '["--config", "sandbox_workspace_write.network_access=true"]'
prompt: |
You are diagnosing a failed Video.js E2E workflow for a pull request.
</file context>
| effort: medium | ||
| sandbox: workspace-write | ||
| safety-strategy: drop-sudo | ||
| codex-args: '["--config", "sandbox_workspace_write.network_access=true"]' |
There was a problem hiding this comment.
P1: When a maintainer reopens an attacker-controlled issue, Codex receives its body as prompt data, inherits the write-capable GH_TOKEN, and can make arbitrary network requests. A prompt injection can therefore exfiltrate the token or mutate issues; run untrusted issue analysis without credentials or network access, then apply only validated results in a separate trusted step.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .github/workflows/issue-triage.yml, line 35:
<comment>When a maintainer reopens an attacker-controlled issue, Codex receives its body as prompt data, inherits the write-capable `GH_TOKEN`, and can make arbitrary network requests. A prompt injection can therefore exfiltrate the token or mutate issues; run untrusted issue analysis without credentials or network access, then apply only validated results in a separate trusted step.</comment>
<file context>
@@ -17,20 +16,23 @@ jobs:
+ effort: medium
+ sandbox: workspace-write
+ safety-strategy: drop-sudo
+ codex-args: '["--config", "sandbox_workspace_write.network_access=true"]'
prompt: |
You are the issue triage agent for this repository.
</file context>
| --model sonnet | ||
| --max-turns 12 | ||
| --allowedTools "Bash(gh:*)" "WebSearch" "WebFetch" | ||
| openai-api-key: ${{ secrets.OPENAI_API_KEY }} |
There was a problem hiding this comment.
P1: When an external user opens or reopens an issue, openai/codex-action@v1 rejects the run before Codex because allow-users defaults to empty and the actor lacks repository write access. This disables the workflow’s external-issue triage path; add the intended external-user allowlist or route this job through a trusted bot.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .github/workflows/issue-triage.yml, line 30:
<comment>When an external user opens or reopens an issue, `openai/codex-action@v1` rejects the run before Codex because `allow-users` defaults to empty and the actor lacks repository write access. This disables the workflow’s external-issue triage path; add the intended external-user allowlist or route this job through a trusted bot.</comment>
<file context>
@@ -17,20 +16,23 @@ jobs:
- --model sonnet
- --max-turns 12
- --allowedTools "Bash(gh:*)" "WebSearch" "WebFetch"
+ openai-api-key: ${{ secrets.OPENAI_API_KEY }}
+ model: gpt-5.6-terra
+ effort: medium
</file context>
| async function render() { | ||
| await prepareSandboxHtmlLocale(); | ||
|
|
||
| const src = withMuxMaxResolution(SOURCES[source].url ?? '', '720p'); |
There was a problem hiding this comment.
P2: When the sandbox source picker selects mp4-1, this renders its .mp4 URL into the HLS-only background-video engine, so playback fails instead of showing a valid source. Restrict the background preset's available sources to type === 'hls' (while retaining the intentional HLS failure cases such as MPEG-TS and unlicensed DRM).
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At apps/sandbox/templates/html-mux-background-video/main.ts, line 29:
<comment>When the sandbox source picker selects `mp4-1`, this renders its `.mp4` URL into the HLS-only background-video engine, so playback fails instead of showing a valid source. Restrict the background preset's available sources to `type === 'hls'` (while retaining the intentional HLS failure cases such as MPEG-TS and unlicensed DRM).</comment>
<file context>
@@ -3,35 +3,45 @@ import { bindSandboxHtmlLocaleChange, prepareSandboxHtmlLocale, wrapSandboxHtmlI
async function render() {
await prepareSandboxHtmlLocale();
+ const src = withMuxMaxResolution(SOURCES[source].url ?? '', '720p');
+
document.getElementById('root')!.innerHTML = wrapSandboxHtmlI18n(html`
</file context>
| "types": "./dist/dev/components.d.ts", | ||
| "development": "./dist/dev/components.js", | ||
| "default": "./dist/default/components.js" | ||
| "./vjsc": { |
There was a problem hiding this comment.
P2: The new "./vjsc" subpath export is only added to the top-level exports, not to publishConfig.exports. npm publishes with the publishConfig.exports object replacing the manifest's exports, so the published @videojs/core will not expose ./vjsc and any consumer importing @videojs/core/vjsc (including packages/skins/vjsc/components/*) will fail to resolve it after publish even though it works in the workspace. Mirror the ./vjsc entry in publishConfig.exports.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At packages/core/package.json, line 38:
<comment>The new "./vjsc" subpath export is only added to the top-level exports, not to publishConfig.exports. npm publishes with the publishConfig.exports object replacing the manifest's exports, so the published @videojs/core will not expose ./vjsc and any consumer importing @videojs/core/vjsc (including packages/skins/vjsc/components/*) will fail to resolve it after publish even though it works in the workspace. Mirror the ./vjsc entry in publishConfig.exports.</comment>
<file context>
@@ -29,10 +35,10 @@
- "types": "./dist/dev/components.d.ts",
- "development": "./dist/dev/components.js",
- "default": "./dist/default/components.js"
+ "./vjsc": {
+ "types": "./dist/dev/vjsc.d.ts",
+ "development": "./dist/dev/vjsc.js",
</file context>
| writeFileSync(output, generate()); | ||
| const generated = generate(); | ||
|
|
||
| if (!existsSync(output) || readFileSync(output, 'utf8') !== generated) writeFileSync(output, generated); |
There was a problem hiding this comment.
P3: The log always prints "Updated generated translation types" even when the new existsSync/readFileSync guard skips the write because the file is already current. In watch/dev runs this reports an update that did not happen. Print only when a write occurred, e.g. move the log inside the condition or log a no-change notice.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At packages/core/scripts/generate-i18n-types.ts, line 38:
<comment>The log always prints "Updated generated translation types" even when the new `existsSync`/`readFileSync` guard skips the write because the file is already current. In watch/dev runs this reports an update that did not happen. Print only when a write occurred, e.g. move the log inside the condition or log a no-change notice.</comment>
<file context>
@@ -30,5 +33,8 @@ ${params}
-writeFileSync(output, generate());
+const generated = generate();
+
+if (!existsSync(output) || readFileSync(output, 'utf8') !== generated) writeFileSync(output, generated);
+
console.log('[generate-i18n-types] Updated generated translation types');
</file context>
|
|
||
| **Child (`<img>`):** User provides appropriate `alt` text. Whether a poster is informative or decorative is the author's judgment (per [WAI guidelines](https://www.w3.org/WAI/tutorials/images/decorative/)). This is an advantage over Media Chrome (which forces `aria-hidden="true"` on the internal image) and native `<video poster>` (which has no `alt` equivalent). | ||
| **Image:** The one each binding renders carries `alt=""`. A resolved URL says nothing about what it depicts, and announcing the URL is worse than announcing nothing. Supply your own `alt` — as a prop in React, on your own `<img>` in HTML — to describe a poster that carries meaning. Whether a poster is informative or decorative is the author's judgment (per [WAI guidelines](https://www.w3.org/WAI/tutorials/images/decorative/)). This is an advantage over Media Chrome (which forces `aria-hidden="true"` on the internal image) and native `<video poster>` (which has no `alt` equivalent). |
There was a problem hiding this comment.
P3: 'The one each binding renders' overclaims for the HTML binding. The HTML element renders no image and only sets src on the author's — it cannot attach alt="". Only the React binding renders an img with alt="". Say the default applies to the image the React binding renders, while the HTML binding fills src only (alt stays the author's), consistent with the following sentences about supplying your own alt.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At internal/design/ui/poster.md, line 31:
<comment>'The one each binding renders' overclaims for the HTML binding. The HTML element renders no image and only sets src on the author's <img> — it cannot attach alt="". Only the React binding renders an img with alt="". Say the default applies to the image the React binding renders, while the HTML binding fills src only (alt stays the author's), consistent with the following sentences about supplying your own alt.</comment>
<file context>
@@ -11,20 +11,24 @@ Display component for video poster image. Shows before playback starts, hides af
+**HTML element (`<media-poster>`):** No ARIA role needed. Custom elements have no implicit role, so there's no semantics to hide or override. Do not add `aria-hidden` — the poster image may be informative.
-**Child (`<img>`):** User provides appropriate `alt` text. Whether a poster is informative or decorative is the author's judgment (per [WAI guidelines](https://www.w3.org/WAI/tutorials/images/decorative/)). This is an advantage over Media Chrome (which forces `aria-hidden="true"` on the internal image) and native `<video poster>` (which has no `alt` equivalent).
+**Image:** The one each binding renders carries `alt=""`. A resolved URL says nothing about what it depicts, and announcing the URL is worse than announcing nothing. Supply your own `alt` — as a prop in React, on your own `<img>` in HTML — to describe a poster that carries meaning. Whether a poster is informative or decorative is the author's judgment (per [WAI guidelines](https://www.w3.org/WAI/tutorials/images/decorative/)). This is an advantage over Media Chrome (which forces `aria-hidden="true"` on the internal image) and native `<video poster>` (which has no `alt` equivalent).
## Alternatives Considered
</file context>
| --max-turns 25 | ||
| --allowedTools "Read" "Glob" "Grep" "Bash(gh:*)" "Bash(cat:*)" "Bash(jq:*)" "Bash(diff:*)" "Bash(ls:*)" "Bash(head:*)" "Bash(tail:*)" "Bash(wc:*)" "Bash(sort:*)" | ||
| openai-api-key: ${{ secrets.OPENAI_API_KEY }} | ||
| model: gpt-5.6-terra |
There was a problem hiding this comment.
P3: This job is labeled GPT-5.6 Sol but selects gpt-5.6-terra, so it runs a different model than the workflow and PR description advertise. Set the model to gpt-5.6-sol or align the label and description with Terra.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .github/workflows/api-reference-sync.yml, line 281:
<comment>This job is labeled GPT-5.6 Sol but selects `gpt-5.6-terra`, so it runs a different model than the workflow and PR description advertise. Set the model to `gpt-5.6-sol` or align the label and description with Terra.</comment>
<file context>
@@ -269,18 +273,24 @@ jobs:
- --max-turns 25
- --allowedTools "Read" "Glob" "Grep" "Bash(gh:*)" "Bash(cat:*)" "Bash(jq:*)" "Bash(diff:*)" "Bash(ls:*)" "Bash(head:*)" "Bash(tail:*)" "Bash(wc:*)" "Bash(sort:*)"
+ openai-api-key: ${{ secrets.OPENAI_API_KEY }}
+ model: gpt-5.6-terra
+ effort: high
+ sandbox: workspace-write
</file context>
| model: gpt-5.6-terra | |
| model: gpt-5.6-sol |
| const player = create(); | ||
| player.play(); // Error! | ||
| const { PlayerElement } = createPlayer({ features: videoFeatures }); | ||
| const player = new PlayerElement(); |
There was a problem hiding this comment.
P3: The corrected example now uses player.append(videoElement) + document.body.append(player) and never calls attach(), but the section's cause still says "Called method before attach()." Update the cause/title to reference the missing media append so the troubleshooting text matches the revised API.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .agents/skills/write-docs/references/state-tooling.md, line 280:
<comment>The corrected example now uses `player.append(videoElement)` + `document.body.append(player)` and never calls `attach()`, but the section's cause still says "Called method before `attach()`." Update the cause/title to reference the missing media append so the troubleshooting text matches the revised API.</comment>
<file context>
@@ -276,15 +276,16 @@ Document common errors:
-const player = create();
-player.play(); // Error!
+const { PlayerElement } = createPlayer({ features: videoFeatures });
+const player = new PlayerElement();
+await player.store.play(); // No media is attached yet.
</file context>
Summary
Sync
mmcc/moqwith upstreamvideojs/v10main atd08395023(10.0.0-beta.32, 165 commits) as a single merge commit, resolving the conflicts with our multi-switching-set track selection and porting the MoQ playback branch onto upstream's new Vite+ toolchain.Reviewing this PR
The "Files changed" tab is the whole upstream delta and is not useful. Review the merge commit's manual work instead — everything beyond git's automatic merge:
Changes
switchVideoTrackkeeps the all-sets candidate pool andconfineToActiveSwitchingSetas the leading constraint, now alongside upstream'sstickToSelectedCodecsconstraint andpreferCodecFamilies/playerResolutionCaprules.moqpack entry moves topackages/spf/vite.config.ts;simple-moq-videostays off the CDN via an exclusion inpackages/html/vite.config.ts(CDN media entries are now auto-discovered fromdefine/media/).vite-plus/test;safeDefinecomes fromregistration/safe-define.anti-sloppadding rule (whitespace/comment reflow only).findTrack,activeVideoTrack,isLiveTrack,measure-size --moq) keep our text, reflowed to upstream's style..agents/plans/(.claude/plansis a generated alias now).Testing
pnpm typecheck,pnpm lint,pnpm check:workspace,pnpm build:packages,pnpm build:sandbox— all greenpnpm -F @videojs/spf test: 119 files / 2055 tests passed (browser projects included)pnpm -F @videojs/html test: 63 files / 480 tests passedNote: the merge commit was made with
--no-verifybecause lint-staged cannot run on merge commits (Needed a single revision); the full repo lint ran green beforehand.🤖 Generated with Claude Code
Summary by cubic
Syncs
mmcc/moqwith upstreamvideojs/v10main at 10.0.0-beta.32 (165 commits) as a single merge commit, porting the MoQ playback branch onto upstream's Vite+ toolchain and resolving conflicts with our multi-switching-set track selection. The Files changed tab is the whole upstream delta and won't help review; review the merge commit's manual work instead viagit show --remerge-diff.Manual merge work
switchVideoTrackkeeps its all-sets candidate pool, now hardened with upstream'sstickToSelectedCodecs,preferCodecFamilies, andplayerResolutionCaprules.moqpack entry moves topackages/spf/vite.config.ts;simple-moq-videostays off the CDN via an exclusion there.vite-plus/test;safeDefinecomes fromregistration/safe-define.anti-slopstyle — whitespace and comment reflow only..agents/plans/(.claude/plansis now a generated alias).Testing
check:workspace, and package/sandbox builds pass;@videojs/spf(2055 tests) and@videojs/html(480 tests) pass.--no-verifybecause lint-staged cannot run on merge commits; the whole repo lint ran green beforehand.Written for commit 30d3001. Summary will update on new commits.