agentleFS
brandonbryant12brandonbryant12/clearframe/skills/clearframe-scene/SKILL.md

clearframe-scene

Skill1 starsChanged 23 days agoMITView on GitHub

Author native GPU stages for ClearFrame films (actors with stable identities, links that follow them, packets, callouts, commit-grounded code edits, footage textures, SkSL materials, particles, a 2.5D camera with depth of field and real motion blur) and develop the scene engine that renders every film. Use for technical/PR explainers, systems that change over time, footage-led shots with native type, material looks, and for engine or compositor changes.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    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. Very specific:

    Keep steady films steady: no camera keys or shutter on a business/report stage unless the brief asks for movement.

    line 26
  2. Very specific:

    Run sheet, check, then render and look at decoded frames around every cut and camera move.

    line 27

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Native stages and the scene engine

Read docs/scene-engine.md (design and full reference) and docs/scene-engine-coverage.md (every block, sketch, treatment and playbook with its measured state). The scene engine is ClearFrame's only renderer: blocks, stages, compositing and footage. Renders run through fframes as the runtime (packages/engine/src/scene/native/src/runtime.rs; docs/fframes-return.md): change pictures in the engine, and scheduling, encoding or segment rules in runtime.rs.

When to use a stage

  • A mechanism with identities that persist: a request travelling between services, a state machine changing, a component replaced. Use actors + links + events instead of re-drawing a diagram per beat. A film stages entry keeps the same actors across several beats.
  • Code that changed: code: {commit, file, window} reads the exact commit from git, keeps unchanged lines in place, opens room for insertions and closes it for removals. The plan records the commit and blob ids; put the commit in source.
  • Footage with native type: video elements are decoded as GPU textures at the size drawn; title them with camera: false text so a camera push does not carry the type off frame.
  • A look: ground or material (noise, sheen, halftone, grain, glass, chrome, gold, thermal, scanlines, dither, neon), particles (burst, stream along a link, field), shutter for motion blur on fast moves.

Charts, KPIs, plots, quotes and kinetic speech stay in their blocks (exact values, sources, audits). Put a stage over a block for callouts and emphasis, not a re-drawn chart.

Authoring rules

  1. Positions are frame pixels of the layout frame (1920×1080 landscape, 1080×1920 vertical; the shorter side is 1080). Keep type inside title-safe; check audits native type like block type. Author tall positions for a vertical film: a wide-authored stage in a tall frame is re-laid down the frame automatically (and check says so), but only in order, not where you would place it. Code editors fit their frame by themselves (type to the longest line, ≥ 24 px, then hanging wraps).
  2. Cue with say (a word of the beat's narration; for film stages the first match in its beats, or {beat, say}). Every cue must land inside the stage; late cues fail.
  3. Actor ids are the identities: links, packets (send), callouts and attach refer to them. Events can be in any order; keys are sorted.
  4. Footage must cover its time on screen, including the time a wipe into the next beat holds it; it never loops, and freezes only with hold: true.
  5. Canvas-only features (rough, print, mosaic, morph, solid, meter, loop: level) are refused on stages by name. Use a canvas beat or art for them.
  6. Keep steady films steady: no camera keys or shutter on a business/report stage unless the brief asks for movement.
  7. Run sheet, check, then render and look at decoded frames around every cut and camera move.

Minimal stage

{ "id": "flow", "block": "stage", "vo": "The CLI sends stop; the daemon waits for the child to exit.",
  "props": { "title": "How a stop request travels", "source": "PR #12, commit 1a2b3c4",
    "actors": [ {"id": "cli", "label": "CLI", "kind": "client", "x": 420, "y": 600},
                {"id": "daemon", "label": "Daemon", "x": 960, "y": 600},
                {"id": "child", "label": "Child", "kind": "process", "x": 1500, "y": 600} ],
    "links": [ {"from": "cli", "to": "daemon", "label": "stop"}, {"from": "daemon", "to": "child", "label": "SIGTERM"} ],
    "events": [ {"do": "send", "from": "cli", "to": "daemon", "say": "sends"},
                {"do": "callout", "actor": "daemon", "text": "waits for exit", "say": "waits"},
                {"do": "state", "actor": "child", "status": "done", "label": "Exited", "say": "exit"} ],
    "shutter": 0.5 } }

Example: examples/stage-pr (a film stage across beats, a callout riding a camera move, a code edit from a commit).

Engine development

  • Sources: packages/engine/src/scene/native/src — plan.rs (contract), compose.rs (frame order, wipes, motion blur), film.rs (backdrop, texture, chrome, frame), nodes.rs (elements, recipes' targets, camera), materials.rs (SkSL), media.rs (footage/images, bounded), fonts.rs (glyph outlines from the rustybuzz shaping that measured them), draw.rs (the blocks' display list and painter), fx.rs (filter graphs), blocks/ (the 31 blocks), encode.rs (FFmpeg pipe), inspect.rs (inspect + audit). JS: packages/engine/src/scene/compile.mts, packages/engine/src/scene/recipes.mts, packages/engine/src/scene/code.mts, packages/engine/src/scene/engine.mts.
  • Frame output is a pure function of the frame and prepared inputs. Anything random is hash-derived from a seed; no wall clock.
  • Measure a change in the picture against earlier stills: scripts/engine-parity.mjs OUT --reference DIR (library) and scripts/engine-effects.mjs OUT --reference DIR (film-level effects). The retained FFFrames-era stills in build/parity and build/engine-effects are such references.
  • Build: pnpm cf build (one Cargo job). Tests: node --test test/scene-engine.test.mjs; Rust: env CARGO_TARGET_DIR=scene/.cache/target cargo test --manifest-path packages/engine/src/scene/native/Cargo.toml --release --locked --jobs 1 -- --test-threads=2.
  • Measure before claiming speed: scripts/engine-bench.mjs OUT PROJECT… (cold, warm, range, full, peak memory, temp disk).

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.