agentleFS
kajisho5kajisho5/color-grading-skill/SKILL.md

color-grading

Skill7 starsChanged 6 days agoMITView on GitHub

Deterministic colour grading / colour correction execution Skill for the AI Video Production Ecosystem. Use it when a caller (normally video-production-agent) has already decided which colour treatment to apply to a video and needs it executed safely - HDR (PQ/HLG, BT.2020) to SDR BT.709 tone mapping with an explicit curve, application of a 3D .cube LUT, rewriting colour tags (BT.709 / BT.2020 PQ / BT.2020 HLG / BT.601) without a re-encode, removing a Dolby Vision RPU, or typed primary colour correction (exposure, contrast, saturation, white balance via temperature + tint, gamma, three-way shadows/highlights via lift + gain, levels, or a curves preset), soft-clipping into broadcast-legal levels (LEGALIZE, BT.709 or EBU R 103; the shares of samples out of range are measured before and after only when ffmpeg-skill has check --levels and the picture is Y'CbCr, otherwise the report sets measured to false), matching one camera to a reference camera from a frame pair the caller names (MATCH_CAMERA), or an OpenColorIO / ACES colour-space transform named by a config (camera log -> ACEScct -> Rec.709 or Rec.2100 PQ; optional PyOpenColorIO extra) - as a typed operation graph with validated outputs and provenance. Do NOT use it to decide which colour treatment, look, LUT or correction values to apply (video-production-agent), to measure or analyse media (media-analysis-skill), to edit video (video-editing-skill), or to run arbitrary ffmpeg commands or filters (it refuses them). WHITE_BALANCE is accepted as an alias of PRIMARY_CORRECTION restricted to temperature/tint (resolved at validation; results report PRIMARY_CORRECTION).

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. notivn/AIEVv2★ 136same name
    2. notivn/AIEV★ 132same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    Never send commands, argv, filter strings or executable paths: the request is rejected.

    line 25
  2. Very specific:

    Keep outputs inside the workspace, never at the input path.

    line 27
  3. Very specific:

    Read results[].status, outputs[].provenance and error from the JSON.

    line 29

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

color-grading

Machine interface: color-grading run - --json with a color-grading/request@1 document on stdin; exactly one color-grading/response@1 document on stdout. skill --json prints the contract, doctor --json the environment, plan - --json a dry run, validate - --json a schema check.

Rules for a calling agent:

  1. Decide first (video-production-agent): which operation, which LUT, which tonemap curve, which retag target. This skill has no defaults for creative parameters that matter (LUT path, retag target) and never guesses which colour treatment to apply.
  2. Give explicit, typed parameters: HDR_TO_SDR needs nothing but has typed tonemap/peak_nits/desat/force (without tonemap the curve follows the input's transfer: mobius for HLG, hable for PQ and anything else, ADR-28); LUT_APPLY needs lut_path; RETAG needs target; PRIMARY_CORRECTION needs nothing but has typed exposure/contrast/saturation/temperature/tint/gamma/lift/gain/levels_in_black/ levels_in_white/levels_out_black/levels_out_white/curves, each independently optional and defaulting to unchanged (curves defaults to no preset); OCIO_TRANSFORM needs source_space and either destination_space or display + view (names from the config), optional config/look/lut_size/output_tags; MATCH_CAMERA needs reference_path (camera A), with at/reference_at, model, optional patches, lut_size. Tone-mapping curve and correction values are always the caller's, never automatic.
  3. Never send commands, argv, filter strings or executable paths: the request is rejected. A LUT path is data, resolved through its own path policy and hashed into provenance; it is never a filter string.
  4. Keep outputs inside the workspace, never at the input path; set overwrite: true deliberately. An output's container must match the source's (this skill does not convert containers; that is ffmpeg-skill/export).
  5. Read results[].status, outputs[].provenance and error from the JSON; the exit code is 0 only when ok is true.
  6. An alpha source (ProRes 4444, qtrle, PNG) keeps its alpha through LUT_APPLY / PRIMARY_CORRECTION when ffmpeg-skill has --keep-alpha; other re-encoding operations refuse it (alpha_not_preserved). Say how its colour is stored with project.source.alpha_mode: straight | premultiplied (unsaid: PNG is straight, anything else is taken as straight with a warning). A premultiplied source is refused for those grades (alpha_premultiplied: ffmpeg-skill cannot un-premultiply it). project.source.drop_alpha: true asks for an opaque result instead (ADR-33).

See README.md for the schemas, operation table, error codes and limitations.

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.