agentleFS
Mathews-TomMathews-Tom/armory/skills/html-video/SKILL.md

html-video

Skill340 starsChanged 7 months agoMITView on GitHub

Use when the user wants a video authored as HTML and CSS and rendered with Hyperframes: "html video", "hyperframes", "render html to mp4", "animate html css to video", "title sequence from html", "html caption video", "side by side comparison video in html", "diagram explainer video in html". NOT for React or Remotion compositions, Manim animations, analysing an existing video, ordinary web pages with no video output, or planning a launch video from the current project (use launch-video, which renders through this package).

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No cloud rendering, avatar or model generation, publishing, or skill installation is part of this workflow.

    line 23
  2. Very specific:

    Never install system binaries yourself, never run Hyperframes through npx, and never use @latest.

    line 29
  3. Distinctive:

    Add no GSAP, no CDN script, no wall-clock timers and no unseeded randomness.

    line 54
  4. Very specific:

    Keep each top-level clip single-level (text, or an element without block children) to keep lint clean.

    line 55

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

HTML Video

Author a video as plain HTML and CSS, then render it with a pinned, project-local Hyperframes CLI. The bundled CSS-only templates use local assets after explicit bootstrap provisioning; authoring and static audit need no network. Rendering arbitrary project code is not an offline sandbox: enforce external network isolation when that guarantee is required. Deliver editable HTML, its statically reachable asset closure, provenance and licence records, the verification reports, and the MP4 together.

When to Use

RequestFit
Titles, comparison, diagram walk-through or captions rendered to MP4 from HTML/CSSUse a bundled template
Edit or re-render an existing Hyperframes compositionAudit it first, then run the workflow
React or Remotion component, Manim scene, existing-video analysis, plain web pageNot this skill

No cloud rendering, avatar or model generation, publishing, or skill installation is part of this workflow. Do not read or invoke any other package from here.

Prerequisites

SKILL_DIR is the absolute directory containing this file. The scripts need Python 3.12+ and only the standard library. Rendering needs Node (major in config.toml), FFmpeg with ffprobe, a Chrome or chrome-headless-shell executable, and a font file the user may redistribute. The Hyperframes version is pinned once, in config.toml; never state or override it elsewhere.

Never install system binaries yourself, never run Hyperframes through npx, and never use @latest. If a prerequisite is missing, stop and tell the user what to install.

python3 "$SKILL_DIR/scripts/html_video.py" --help
python3 "$SKILL_DIR/scripts/html_video.py" init --list

Workflow

  1. Pick and copy a template. init TEMPLATE DEST copies title-sequence, comparison, diagram-explainer or captioned-sequence. All are 1920x1080, text-editable and animated by finite CSS keyframes.
  2. Get the font from the user. Ask for a font file (woff2, woff, ttf or otf), its licence text, SPDX id and source. Never download or assume one; never bundle a system font. fonts add PROJECT --font-file F --family NAME --license-id ID --license-file L --source S copies it, records provenance, writes fonts.css in the project's assets directory, and updates compiler-visible inline @font-face declarations in index.html. Templates use the alias HV Text; no manual HTML edit is needed. --family is only the provenance name; the CSS family is --alias (default HV Text). Provision a second family, such as a monospace, with its own --alias "HV Code" on each of its faces and reference that alias in CSS, or it becomes another face of HV Text.
  3. Bootstrap once. bootstrap PROJECT --browser-path PATH pins an existing browser, or --download-browser runs the pinned CLI's own browser ensure. The package manager comes from the project lockfile (npm for a fresh project). The lock .html-video/environment.json records tool paths and actual versions, including the browser's own --version. Afterwards doctor PROJECT re-verifies offline.
  4. Author. Edit text and data-start / data-duration values; see the contract below. Keep the root data-duration equal to the last clip end.
  5. Audit. audit PROJECT rejects remote render-critical resources, path and symlink escapes, missing files, case mismatches, cycles, undeclared fonts and invalid timing. Fix every error.
  6. Verify. run lint PROJECT, then run check PROJECT, then run snapshot PROJECT. Read the coverage table below, then open every .html-video/snapshots/contact-sheet*.jpg page and judge the frames yourself: layout, legibility, and motion at the sampled boundaries.
  7. Render. run render PROJECT uses the FPS from config.toml, renders with strict, non-best-effort, software-browser capture and one worker, removes any previous output first, and checks the MP4 against the composition. Override FPS only by passing the same --fps N to all four run commands.
  8. Deliver. deliver PROJECT --out DIR requires passing reports for the same source, verified toolchain, config and effective FPS, verifies every recorded artifact, re-audits copied source/ with provenance required, and writes RENDER.json plus reports/. Pass --fps N here too after an FPS override. Any failed rerun invalidates its prior report.

plan PROJECT prints the frame-aligned boundary times used for check and snapshots.

Timing and motion contract

  • The root has data-composition-id, data-start="0", integer data-width/data-height, a finite positive data-duration, and data-no-timeline when no timeline is registered.
  • Each clip has a unique id, data-start (a number, or otherId [+/- seconds]) and data-duration. Paint order is CSS z-index, not track index.
  • Motion is finite CSS keyframes on HTML elements with animation-fill-mode: both. The runtime seeks them to the clip-local time t - data-start. SVG child elements are not seeked, CSS transitions are not seeked, and infinite iteration is rejected.
  • Add no GSAP, no CDN script, no wall-clock timers and no unseeded randomness. Seeking can land on any time in any order.
  • Keep each top-level clip single-level (text, or an element without block children) to keep lint clean; caption cues use class caption-cue on one track. A zero-height absolutely positioned box can stretch; give shapes real dimensions or use clip-path.

Coverage you must read

run check passes --at boundary times and then enforces these report fields. A pass without them proves nothing.

FieldRequiredMeaning
browserSkippedfalseBrowser-backed layout, motion and contrast ran
layout.samplesevery requested timeBoundaries were really sampled
motion.enabled, motion.samplestrue, > 0 when animatedNeeds the index.motion.json sidecar
contrast.checked> 0 when text existsAbsent text means no contrast proof
_meta.versionthe pinned versionThe pinned CLI produced the report

--at-transitions reads GSAP timelines only, so CSS-animated compositions rely on the explicit plan times.

Limits and network behaviour

  • The audit is a static reference audit, not a sandbox. It follows supported static HTML/CSS/SVG references, local module imports and import-map addresses, but computed URLs and arbitrary JavaScript remain executable. Recognised runtime loading, wall-clock APIs and unseeked Web Animations produce warnings, not proof of their absence. Review every warning; enforce network isolation outside the tool for a hard guarantee.
  • Source-confirmed network paths addressed by the package: telemetry and update checks (forced off), snapshot --describe (the vision key is scrubbed and --describe false is passed), and compile-time Google Fonts resolution (used named families require compiler-visible inline local @font-face). Known unmapped-font or Google Fonts compiler logs fail the run rather than silently falling back.
  • The browser is pinned by passing its path through HYPERFRAMES_BROWSER_PATH and PRODUCER_HEADLESS_SHELL_PATH, so a render cannot silently select or download another build. A failed --version or changed version aborts the run.
  • Rendering is seek-driven, but no byte-identical output is claimed. Results depend on the recorded Hyperframes, browser, FFmpeg and operating-system versions and on the font files.

Error Handling

SymptomCauseFix
font_not_declaredA named family has no local @font-faceRun fonts add; never let the compiler fetch it
font_face_not_inline, font_dynamic, font_face_local_sourceLinked-only, scoped-variable or system-dependent font resolutionRun fonts add; use a literal provisioned family and inline local font faces, never local()
remote_resource, path_escape, symlink_escapeResource not inside the projectCopy the file into the project and reference it relatively
motion_spec_missing or motion_unprovenAnimated, no motion sidecarAdd assertions to index.motion.json
contrast_unprovenText present but no pair checkedRe-run check at times where text is visible
report_stale, report_*_mismatch at deliverSource, toolchain, config or FPS changed after verificationRe-run all four steps with one matching context; re-bootstrap only when intentionally changing the toolchain
not installed, changed since bootstrapToolchain differs from the lockRe-run bootstrap; do not fall back to npx
render_mismatchMP4 size, rate or length differsFix data-duration/dimensions and re-render

Output

Report: the delivery directory, the MP4 path, the RENDER.json environment summary (versions, fps, duration), the check coverage numbers you read, which frames you inspected and what you judged, every warning left open, and the font and asset licences declared by the user.

Reference

  • references/composition-contract.md for the verified runtime seek contract and authoring rules
  • references/fonts-and-assets.md for font provisioning, provenance schema and what the audit proves
  • references/verification.md for command flags, coverage fields and the reproducibility record
  • NOTICE.md for attribution and terms

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.