[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7FCNG59KTE2AXGCER1VR5AJ":3,"skill-same:01M7FCNG59KTE2AXGCER1VR5AJ":373},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":31,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M7FCNG59KTE2AXGCER1VR5AJ","taste-application",null,"Generate new video against a distilled style pack and cut it into a finished piece - plan takes from the reference's cut rhythm, generate on fal, grade with the pack's measured LUT, cut at the measured cadence, weave in existing footage, composite overlay plates, mint 3D props, and verify the result numerically. Use when the user wants to make a video in a captured style, supplement existing footage, or assemble generated clips into a real edit.","siddu1004\u002FSFP","siddu1004","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F260113491?v=4","01M7F8W1SAJAR49097VRTE1TW5","Siddhardh Vanguri","01M7F8W1SYGY7JE15AHAECCT9R",5,"MIT","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: taste-application\ndescription: Generate new video against a distilled style pack and cut it into a finished piece - plan takes from the reference's cut rhythm, generate on fal, grade with the pack's measured LUT, cut at the measured cadence, weave in existing footage, composite overlay plates, mint 3D props, and verify the result numerically. Use when the user wants to make a video in a captured style, supplement existing footage, or assemble generated clips into a real edit.\nmetadata:\n  origin: ECC\n---\n\n# Taste Application\n\nThe second half of the pipeline. **taste-distillation** measures references into\na style pack; this generates against that pack and cuts the result.\n\n## Execution and Delivery Contract\n\nThe original implementation ships here in `scripts\u002F`; no separate `ito-video`\ncheckout is required. Install `scripts\u002Frequirements.txt` for local processing.\nInstall `scripts\u002Frequirements-live.txt` only for provider execution. Live\nuploads, submissions and downloads require explicit `TASTE_FORGE_ALLOW_LIVE=1`\nin addition to credentials; set it only for the user's authorized run.\n`--dry-run` remains credential-free and produces labelled placeholders.\nAn ambiguous provider timeout is not retried as a new paid job. Inspect the\nprovider request before deciding whether another submission is warranted.\n\nUse existing completed takes without any provider calls:\n\n```bash\npython scripts\u002Fpipeline.py --genre example --root stylepacks \\\n  --takes media\u002Ftake-a.mp4 media\u002Ftake-b.mp4 --duration 12 --fps 30 \\\n  --out out\u002Freview-v1.mp4\n```\n\n`--duration` is a **best-effort cadence target**, not an exact runtime. Complete\nshots may produce a shorter or longer edit; the assembler does not duplicate\nclips or add padding to meet the target. The output manifest retains actual\n`duration` and adds `duration_contract` with requested and actual seconds,\nshortfall, overrun, and the `cadence_target` policy. Differences of at least\none output frame are warned explicitly. No exact-duration mode is provided;\nwhen an exact runtime is required, inspect the receipt and revise or reject\nthe cut before delivery.\n\nThe pipeline keeps each run's graded shot files because the editable FCPXML\nand EDL reference them. It refuses output collisions and reports timeline\nexport failures. A completed render is not a saved editor project or creative\napproval. Preserve source assets and versioned project checkpoints before\nand after live edits; record the saved path and digest separately from the\nin-memory timeline receipt.\n\nClone-ready Fal graphs and their offline input compiler are documented in\n`workflows\u002FREADME.md`. Compile brief and style direction before submission;\ndo not assume a disconnected schema field affects a model prompt. Validate\ncurrent endpoint fields against the actual provider before a paid run.\n\nFor Blender, use the full textured source GLB; retopology is a separately\nnamed derivative and never replaces that source. `blender_prop.py` supports\nexplicit `--width 1920 --height 1080 --fps 30 --receipt receipt.json` and\npreserves pack-derived rim lighting and material textures. Saving a scene is\ndistinct from rendering it; inspect the receipt's state and packed images.\n\nFor Resolve overlays, use `taste.resolve.apply_placements` with injected\nobjects from an explicitly selected, versioned target. Set `source_end_mode`\nto the convention verified on that host. Do not assume an inclusive source\nend across Resolve versions. The adapter allocates overlapping effects above\npreserved tracks and checks every placement immediately and again after all\nappends. Composite integers must match the installed API; for the verified\nStudio 21 host, Screen is 5, not the historical builder's incorrect 22.\nThe receipt proves in-memory placement only. Save the project, verify its\ncheckpoint, then inspect the exact rendered output before reporting delivery.\n\n## When to Activate\n\n- \"make a video in this style\" \u002F \"apply the pack\" \u002F \"supplement this footage\"\n- Assembling generated clips into something with real edit rhythm\n- Minting 3D props from a look and getting them back into the video\n- Verifying that a finished piece actually matches its reference\n\n## Division of Labour\n\n**The model supplies content, motion, framing and lighting structure. The pack\nsupplies colour and rhythm.** This is measured, not stylistic preference — see\ntaste-distillation for the numbers. Practical consequences:\n\n- The generation prompt contains **zero colour language**. Add\n  *\"Colour: none. Render neutral. Grading is applied afterwards.\"*\n- Keep **two separate flags**: `--brief` (what HAPPENS: subject, action, place)\n  and `--style-steer` (how it LOOKS). Merging them leaks style words into the\n  scene (\"teal\" becomes a teal object) and subject words into the grade.\n- The model responds to **local, checkable rules** far better than global ones.\n  \"Backgrounds pure black and unlit; subjects blowing toward white\" works;\n  \"extreme contrast\" does not.\n\n## Generate TAKES, Not Shots\n\nThe obvious reading of \"match the cadence\" is one generation per shot. It is\neconomically absurd. A reference averaging 0.78s\u002Fshot against an endpoint with a\n4-second floor turns a 10s piece into **12 calls, 48 generated seconds for 10\nused (21% efficiency)**, and twelve unrelated clips stitched into what should\nread as continuous.\n\nEditors roll a longer take and cut inside it. Grouping shots into ~5s takes:\n**3 calls, 13 generated seconds, 77% efficiency**, and consecutive shots that\nactually belong to each other because they came from the same generation.\n\nTell the model what shape you want, or it renders a slow locked-off push and six\ncuts inside it read as a stutter:\n\n> \"Filmed as ONE continuous take with no hard cuts inside it. It will be cut into\n> 6 pieces of roughly 0.8s in the edit, so the framing, subject and light must\n> keep changing throughout — any 0.8s window has to stand alone as its own shot.\"\n\n## Cutting Rules\n\n- **Re-encode, never stream-copy.** Stream copy only cuts on keyframes, which at\n  0.78s mean shot length rounds every boundary to the nearest GOP — destroying\n  the exact thing the pipeline exists to preserve.\n- **When supplementing existing footage, use its own shot boundaries.** Slicing a\n  base video into contiguous pieces and playing them in order just reassembles\n  the original: every \"cut\" lands mid-shot and is invisible. Measured, a 20-shot\n  assembly registered only 12 detected cuts. Detect real boundaries and take\n  every Nth so consecutive picks are guaranteed discontinuous. That moved a cut\n  measurement from 1.29s to 0.83s against a 0.78s target.\n- **Sample shot lengths from the reference's distribution**, not from its mean,\n  so the cut inherits rhythm variance instead of flattening to even clips.\n\n## Grading Rules\n\n- **Direct measurement beats a baked LUT** when you have the clip: measure it,\n  match its L\\* CDF, apply zone chroma. `grade_clip_direct` reached MAE 1.58 and\n  contrast 34.0 against 34.7.\n- **Anchor, do not CDF-match, when the clip's histogram is unlike the\n  reference's.** Forcing a 68%-black generated clip onto a busy reference\n  histogram lifted the entire background out of black: background preservation\n  fell to 26.0% (CDF) versus 69.0% (anchor), while MAE and contrast both still\n  looked excellent. Default to anchored tone.\n- **Batch size matters.** Grading 48 frames at once OOM-killed the process;\n  6 is safe.\n\n## Recovered fal Platform Behavior\n\nThese observations and endpoint examples came from the recovered workflow.\nRecheck current endpoint metadata; they are not guarantees about every future\nprovider version. The bundled graph templates record the separately verified\nworkflow inputs, including explicit generated-audio control.\n\nThese cost real time to discover. Check them before designing a graph.\n\n| Limit | Detail |\n|---|---|\n| **No 3D renderer at all** | fal has `image-to-3d`, `text-to-3d`, `3d-to-3d` and nothing else. Every `3d-to-3d` endpoint emits another mesh. There is no `3d-to-image`\u002F`3d-to-video` category, so a minted GLB **cannot** re-enter a fal video graph. Render locally, then use `fal-ai\u002Fffmpeg-api\u002Fimages-to-video`. |\n| **compose cannot overlay** | `fal-ai\u002Fffmpeg-api\u002Fcompose` rejects a second track with *\"Multiple video tracks are not supported\"* — and it counts an `image` track as a video track. It sequences one video track only. **Composite locally with ffmpeg.** |\n| **compose keyframes are milliseconds** | Nothing in the response says so. A run submitted in seconds is accepted and returns a video that is 1000x too short. |\n| **extract-frame offers first\u002Fmiddle\u002Flast only** | No arbitrary timestamp. Use the three as three distinct conditioning images. |\n| **No loops, no string concat in the DAG** | Per-shot fan-out has to be authored node by node, or kept local. |\n| **Kling 3.0 has no reference-to-video** | The v3 line is text\u002Fimage\u002Fmotion-control only; reference-to-video lives on the `o3` line: `fal-ai\u002Fkling-video\u002Fo3\u002Fpro\u002Freference-to-video`. |\n| **Prefixes are not uniform** | `bytedance\u002F*`, `tripo3d\u002F*`, `meshy\u002F*`, `minimax\u002F*`, `openai\u002F*` carry **no** `fal-ai\u002F` prefix. `kling-video`, `veo3.1`, `flux-*`, `hunyuan-3d`, `ffmpeg-api` do. |\n\n### Endpoint picks\n\n| Slot | Best | Value alternative |\n|---|---|---|\n| reference→video | `bytedance\u002Fseedance-2.5\u002Freference-to-video` (~$0.473\u002Fs @720p) | `fal-ai\u002Fkling-video\u002Fo3\u002Fpro\u002Freference-to-video` (~$0.112\u002Fs) |\n| image→3D | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fpro\u002Fimage-to-3d` ($0.375, up to 8 views) | `tripo3d\u002Fh3.1\u002Fimage-to-3d` ($0.20) |\n| text→3D | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fpro\u002Ftext-to-3d` | `tripo3d\u002Fh3.1\u002Ftext-to-3d` |\n| retopology | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fsmart-topology` ($0.75) | `tripo3d\u002Ftripo\u002Fremesh` (~75x cheaper) |\n| part split | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fpart` (FBX only) | `tripo3d\u002Ftripo\u002Fsegment` |\n| text→image | `fal-ai\u002Fnano-banana-pro` ($0.15 flat) | `fal-ai\u002Fflux-2-pro` ($0.03\u002FMP) |\n\nSeedance is ~4x Kling o3's price for the same 5 seconds. It earns that on\nmulti-reference fidelity (up to 50 mixed image\u002Fvideo\u002Faudio refs) and does **not**\nearn it when conditioning on a single still.\n\nModel IDs and prices drift. Verify against fal.ai\u002Fmodels before promising any of\nthem.\n\n## The 3D Branch\n\n```bash\npython mint3d.py --genre \u003Cname> --from-stills 4 --retopo --render\npython mint3d.py --genre \u003Cname> --prompt \"a cracked chrome visor\" --render\n```\n\n- **Generate a clean plate first; do not lift from reference stills.** The\n  endpoint's stated input requirement is simple background, single object,\n  object >50% of frame. Reference reels are the opposite of that — collages,\n  wide shots, several subjects, burnt-in graphics — and they produce sculpted\n  noise. Text → single-object plate → mesh costs ~$0.15 extra and is the\n  difference between a usable mesh and a discarded one.\n- **Multi-view is named per-angle fields, not a list.** `input_image_url` (front,\n  required), then `back_image_url`, `left_image_url`, `right_image_url`,\n  `left_front_image_url`, `right_front_image_url`, `top_image_url`,\n  `bottom_image_url`. There is no `input_image_urls` and no `multi_view` flag —\n  inventing them degrades every mint to single-view while appearing to work. A\n  wrong angle label is worse than omitting the view, because the model trusts it.\n- **Address the GLB by key, not by position.** The response carries a `thumbnail`\n  PNG and a `model_urls` block alongside `model_glb`; taking the first URL works\n  only until the keys reorder, and a preview PNG downloads fine — nothing fails\n  until Blender refuses to open it.\n- **Request PBR maps.** Without them the mesh lights like painted cardboard.\n- **Retopologise** if anyone will edit or rig it. Generated meshes are dense and\n  chaotic.\n- **Render locally to close the loop.** Once a turntable is frames, it is\n  footage, and every downstream stage already handles footage — grade it, cut it,\n  screen it as an element, or upload it as a conditioning reference. Use Blender\n  when a binary is on PATH; keep a dependency-light software rasteriser as the\n  default, because a headless GL context is the single most common thing missing\n  from a container and a renderer that only works on a workstation is not part of\n  a pipeline.\n\n## Verify, Then Believe\n\nEvery other stage claims a result. Check it, and check **distribution shape**,\nnot just moments:\n\n- `background` — share of frame below L\\*10 vs the **pack's** figure. Compare to\n  the reference, **not** to the source clip: a generated source at 68% black is\n  blacker than any reference in a 24–55% band, so \"preserve the source's blacks\"\n  demands the wrong thing and equally excuses a lifted grade.\n- `chroma_mae` — per-zone a\\*\u002Fb\\* error, using the **median** (matching how the\n  pack's targets were measured; a mean here compares a skew-sensitive statistic\n  to a robust one and reports a definition mismatch as an error).\n- `contrast` \u002F `black_point` \u002F `white_point`\n- `banding` — empty L\\* histogram bins *between occupied ones*. Counting total\n  empty bins does not work: a legitimately dark clip has empty highlight bins.\n- `cadence` — detected mean shot length vs the reference's.\n\nWatch the **units trap**: OpenCV changes Lab convention with dtype. On float32,\nL\\* is 0–100 and a\\*\u002Fb\\* are signed; on uint8, L\\* is 0–255 and a\\*\u002Fb\\* are\nbiased +128. Mixing them reports chroma errors in the hundreds.\n\n## Full Chain\n\n```bash\npython pipeline.py --genre \u003Cname> --refs a.mov b.mov \\\n  --brief \"what happens\" --duration 12 \\\n  --base-video existing.mp4 --out out\u002FFINAL.mp4\n```\n\nmint → distill → (mint3d) → apply → forge → verify. Stages 1–3 are cached, so\niterating on briefs never re-measures anything. `--dry-run` stubs every network\ncall: the plan, prompts, track layout and manifest all still get exercised.\n\n## Anti-Patterns\n\n| Don't | Why |\n|---|---|\n| One generation per shot | 21% efficiency, 12 unrelated clips |\n| Put colour in the prompt | Measured not to work; pushes away from the neutral base the LUT wants |\n| Stream-copy the cuts | Keyframe-only boundaries destroy sub-second rhythm |\n| Cut a base video contiguously | Reassembles the original; every cut invisible |\n| Trust compose to overlay | It cannot; it rejects the second track |\n| Send seconds to compose | Silently 1000x too short |\n| Ship on MAE alone | Add the background-share check |\n\n## Borrowed Footage Carries the Capture App's UI\n\nThe single worst defect found in a delivered cut: the finished video shipped\nsomeone else's like button, view counter and comment bubble, because the base\nfootage was a screen recording and nothing cropped them out.\n\n**`content_mask` does not solve this and its bounding box makes it worse.**\nTemporal variance keeps interface chrome, because chrome *animates* - the heart\npulses, the counter ticks - so the mask marks it as moving content. Measured on\nthree references, the mask bbox kept 100% of the width every time while the\ninterface sat plainly in the right-hand margin.\n\nThe separating signal is the temporal **median**, not the variance. Footage\nmoves, so the median of many frames averages into mush with almost no edge\nenergy; chrome sits at fixed coordinates, so its edges survive intact. Sobel\nenergy on the median frame lights up on chrome and goes quiet on content -\nmeasured, a right-hand column read 0.23 against an interior background of 0.03\non one reference and 0.31 against 0.15 on another.\n\nTwo implementation details that cost a cycle each:\n\n- **Trim past the innermost outlier in each outer band, not inward from the\n  edge.** Walking in while the current line is hot stops immediately, because\n  the outermost lines are letterbox - flat black, zero edge energy - and the\n  chrome sits *inside* that at 90-95% of width. The naive version trimmed 1%\n  of frame while the like button stayed in shot.\n- **Scale to cover, not pad,** when portrait source lands in a landscape cut.\n  Padding 9:16 (narrower still after the UI crop) into 16:9 left ~60% of frame\n  as black bars, one shot was nearly an empty rectangle, and it poisoned the\n  background metric because bars are pure black. Covering loses the sides,\n  which is the right trade for centre-framed material.\n\n## Overlay Plates Are Elements, Not Washes\n\nA glow plate is ~4% covered by construction. Composite it at frame size and you\nget a small bright dot parked mid-shot - it reads as a sticker, and it was\nvisible in a delivered cut as an unexplained coloured blob.\n\n- **Tighten each plate to its alpha bounding box first.** That raises coverage\n  from ~4% to 15-35% and hands size control to the caller instead of inheriting\n  whatever fraction of the source frame the element happened to occupy.\n- **Choose wash vs element by coverage.** Diffuse plates (\u003C10% after tightening)\n  work stretched full-frame at low opacity; concentrated ones want to be scaled\n  to 35-70% of frame width and placed.\n- **Vary placement, scale and rotation per shot** from a seeded RNG, so the cut\n  stays reproducible but no two stamped shots share a mark. One plate in one\n  spot every Nth shot reads as a watermark.\n- Resolve element geometry in Python, not in ffmpeg expressions: `pad()` rejects\n  a negative offset and cannot pad below its input size, so an oversized or\n  off-frame element kills the whole filtergraph.\n\n## Downstream Handoff (Resolve \u002F Blender)\n\n**Always ship an editable timeline beside the mp4.** The flattened video is a\nviewing copy and the one thing a colourist cannot work with — every cut is baked\nin and the shots are no longer separable. `forge.py` writes FCPXML 1.9 and a\nCMX3600 EDL referencing the individual graded shot files, so the piece lands as a\ntimeline that can be re-cut and re-graded.\n\nValidate the export, do not assume it: check that asset-clip offsets equal the\nrunning sum of prior durations (no gaps), that every `ref` resolves to a declared\nasset, that every `media-rep src` exists on disk, and that the total matches the\nmp4. A timeline that imports but drifts is worse than one that fails loudly.\n\nTwo things that silently destroy the work:\n\n- **Project frame rate must be set before import.** Resolve locks timeline fps on\n  first timeline creation and conforms the cadence silently. At a sub-second mean\n  shot length that conform is visible.\n- **The delivered shots are already graded.** `look.cube` is a *normalising* LUT\n  for new material and for matching — applying it to the supplied shots\n  double-grades them. Node 1, nothing before it, corrections after.\n\nShip a handoff doc with the measured targets in it (`scripts\u002FHANDOFF-TEMPLATE.md`\nis a filled example): the zone chroma table, the cadence distribution including\n**rhythm variance** — an editor who matches the mean but not the variance\nproduces something that reads completely differently — and an explicit \"do not\"\nlist.\n\nFor Blender, `scripts\u002Fblender_prop.py` derives its lighting from the pack:\nblack world with transparent film (so renders composite with no keying), key plus\nrim (a key alone lets the silhouette die against black), rim colour converted\nfrom the pack's peak-chroma zone via Lab→linear sRGB so the prop picks up the\nsame cast the footage is graded to. **View transform Standard, not AgX\u002FFilmic,\nand no grading in Blender** — AgX applies its own tone curve before the LUT ever\nsees the pixels, and grading twice compounds.\n\n## Bundled Code\n\n`scripts\u002F` in this skill is a working implementation, not pseudocode. It has no\nproject-specific assumptions: point it at any reference videos and it produces a\npack.\n\n```bash\npip install -r scripts\u002Frequirements.txt\nexport FAL_KEY=...            # only needed for the stages that call fal\n```\n\nEvery network call is stubbed under `TASTE_FORGE_DRY_RUN=1` or `--dry-run`, so\nthe plan, prompts, track layout and manifest can be inspected without spending.\n","# Taste Application\n\nThe second half of the pipeline. **taste-distillation** measures references into\na style pack; this generates against that pack and cuts the result.\n\n## Execution and Delivery Contract\n\nThe original implementation ships here in `scripts\u002F`; no separate `ito-video`\ncheckout is required. Install `scripts\u002Frequirements.txt` for local processing.\nInstall `scripts\u002Frequirements-live.txt` only for provider execution. Live\nuploads, submissions and downloads require explicit `TASTE_FORGE_ALLOW_LIVE=1`\nin addition to credentials; set it only for the user's authorized run.\n`--dry-run` remains credential-free and produces labelled placeholders.\nAn ambiguous provider timeout is not retried as a new paid job. Inspect the\nprovider request before deciding whether another submission is warranted.\n\nUse existing completed takes without any provider calls:\n\n```bash\npython scripts\u002Fpipeline.py --genre example --root stylepacks \\\n  --takes media\u002Ftake-a.mp4 media\u002Ftake-b.mp4 --duration 12 --fps 30 \\\n  --out out\u002Freview-v1.mp4\n```\n\n`--duration` is a **best-effort cadence target**, not an exact runtime. Complete\nshots may produce a shorter or longer edit; the assembler does not duplicate\nclips or add padding to meet the target. The output manifest retains actual\n`duration` and adds `duration_contract` with requested and actual seconds,\nshortfall, overrun, and the `cadence_target` policy. Differences of at least\none output frame are warned explicitly. No exact-duration mode is provided;\nwhen an exact runtime is required, inspect the receipt and revise or reject\nthe cut before delivery.\n\nThe pipeline keeps each run's graded shot files because the editable FCPXML\nand EDL reference them. It refuses output collisions and reports timeline\nexport failures. A completed render is not a saved editor project or creative\napproval. Preserve source assets and versioned project checkpoints before\nand after live edits; record the saved path and digest separately from the\nin-memory timeline receipt.\n\nClone-ready Fal graphs and their offline input compiler are documented in\n`workflows\u002FREADME.md`. Compile brief and style direction before submission;\ndo not assume a disconnected schema field affects a model prompt. Validate\ncurrent endpoint fields against the actual provider before a paid run.\n\nFor Blender, use the full textured source GLB; retopology is a separately\nnamed derivative and never replaces that source. `blender_prop.py` supports\nexplicit `--width 1920 --height 1080 --fps 30 --receipt receipt.json` and\npreserves pack-derived rim lighting and material textures. Saving a scene is\ndistinct from rendering it; inspect the receipt's state and packed images.\n\nFor Resolve overlays, use `taste.resolve.apply_placements` with injected\nobjects from an explicitly selected, versioned target. Set `source_end_mode`\nto the convention verified on that host. Do not assume an inclusive source\nend across Resolve versions. The adapter allocates overlapping effects above\npreserved tracks and checks every placement immediately and again after all\nappends. Composite integers must match the installed API; for the verified\nStudio 21 host, Screen is 5, not the historical builder's incorrect 22.\nThe receipt proves in-memory placement only. Save the project, verify its\ncheckpoint, then inspect the exact rendered output before reporting delivery.\n\n## When to Activate\n\n- \"make a video in this style\" \u002F \"apply the pack\" \u002F \"supplement this footage\"\n- Assembling generated clips into something with real edit rhythm\n- Minting 3D props from a look and getting them back into the video\n- Verifying that a finished piece actually matches its reference\n\n## Division of Labour\n\n**The model supplies content, motion, framing and lighting structure. The pack\nsupplies colour and rhythm.** This is measured, not stylistic preference — see\ntaste-distillation for the numbers. Practical consequences:\n\n- The generation prompt contains **zero colour language**. Add\n  *\"Colour: none. Render neutral. Grading is applied afterwards.\"*\n- Keep **two separate flags**: `--brief` (what HAPPENS: subject, action, place)\n  and `--style-steer` (how it LOOKS). Merging them leaks style words into the\n  scene (\"teal\" becomes a teal object) and subject words into the grade.\n- The model responds to **local, checkable rules** far better than global ones.\n  \"Backgrounds pure black and unlit; subjects blowing toward white\" works;\n  \"extreme contrast\" does not.\n\n## Generate TAKES, Not Shots\n\nThe obvious reading of \"match the cadence\" is one generation per shot. It is\neconomically absurd. A reference averaging 0.78s\u002Fshot against an endpoint with a\n4-second floor turns a 10s piece into **12 calls, 48 generated seconds for 10\nused (21% efficiency)**, and twelve unrelated clips stitched into what should\nread as continuous.\n\nEditors roll a longer take and cut inside it. Grouping shots into ~5s takes:\n**3 calls, 13 generated seconds, 77% efficiency**, and consecutive shots that\nactually belong to each other because they came from the same generation.\n\nTell the model what shape you want, or it renders a slow locked-off push and six\ncuts inside it read as a stutter:\n\n> \"Filmed as ONE continuous take with no hard cuts inside it. It will be cut into\n> 6 pieces of roughly 0.8s in the edit, so the framing, subject and light must\n> keep changing throughout — any 0.8s window has to stand alone as its own shot.\"\n\n## Cutting Rules\n\n- **Re-encode, never stream-copy.** Stream copy only cuts on keyframes, which at\n  0.78s mean shot length rounds every boundary to the nearest GOP — destroying\n  the exact thing the pipeline exists to preserve.\n- **When supplementing existing footage, use its own shot boundaries.** Slicing a\n  base video into contiguous pieces and playing them in order just reassembles\n  the original: every \"cut\" lands mid-shot and is invisible. Measured, a 20-shot\n  assembly registered only 12 detected cuts. Detect real boundaries and take\n  every Nth so consecutive picks are guaranteed discontinuous. That moved a cut\n  measurement from 1.29s to 0.83s against a 0.78s target.\n- **Sample shot lengths from the reference's distribution**, not from its mean,\n  so the cut inherits rhythm variance instead of flattening to even clips.\n\n## Grading Rules\n\n- **Direct measurement beats a baked LUT** when you have the clip: measure it,\n  match its L\\* CDF, apply zone chroma. `grade_clip_direct` reached MAE 1.58 and\n  contrast 34.0 against 34.7.\n- **Anchor, do not CDF-match, when the clip's histogram is unlike the\n  reference's.** Forcing a 68%-black generated clip onto a busy reference\n  histogram lifted the entire background out of black: background preservation\n  fell to 26.0% (CDF) versus 69.0% (anchor), while MAE and contrast both still\n  looked excellent. Default to anchored tone.\n- **Batch size matters.** Grading 48 frames at once OOM-killed the process;\n  6 is safe.\n\n## Recovered fal Platform Behavior\n\nThese observations and endpoint examples came from the recovered workflow.\nRecheck current endpoint metadata; they are not guarantees about every future\nprovider version. The bundled graph templates record the separately verified\nworkflow inputs, including explicit generated-audio control.\n\nThese cost real time to discover. Check them before designing a graph.\n\n| Limit | Detail |\n|---|---|\n| **No 3D renderer at all** | fal has `image-to-3d`, `text-to-3d`, `3d-to-3d` and nothing else. Every `3d-to-3d` endpoint emits another mesh. There is no `3d-to-image`\u002F`3d-to-video` category, so a minted GLB **cannot** re-enter a fal video graph. Render locally, then use `fal-ai\u002Fffmpeg-api\u002Fimages-to-video`. |\n| **compose cannot overlay** | `fal-ai\u002Fffmpeg-api\u002Fcompose` rejects a second track with *\"Multiple video tracks are not supported\"* — and it counts an `image` track as a video track. It sequences one video track only. **Composite locally with ffmpeg.** |\n| **compose keyframes are milliseconds** | Nothing in the response says so. A run submitted in seconds is accepted and returns a video that is 1000x too short. |\n| **extract-frame offers first\u002Fmiddle\u002Flast only** | No arbitrary timestamp. Use the three as three distinct conditioning images. |\n| **No loops, no string concat in the DAG** | Per-shot fan-out has to be authored node by node, or kept local. |\n| **Kling 3.0 has no reference-to-video** | The v3 line is text\u002Fimage\u002Fmotion-control only; reference-to-video lives on the `o3` line: `fal-ai\u002Fkling-video\u002Fo3\u002Fpro\u002Freference-to-video`. |\n| **Prefixes are not uniform** | `bytedance\u002F*`, `tripo3d\u002F*`, `meshy\u002F*`, `minimax\u002F*`, `openai\u002F*` carry **no** `fal-ai\u002F` prefix. `kling-video`, `veo3.1`, `flux-*`, `hunyuan-3d`, `ffmpeg-api` do. |\n\n### Endpoint picks\n\n| Slot | Best | Value alternative |\n|---|---|---|\n| reference→video | `bytedance\u002Fseedance-2.5\u002Freference-to-video` (~$0.473\u002Fs @720p) | `fal-ai\u002Fkling-video\u002Fo3\u002Fpro\u002Freference-to-video` (~$0.112\u002Fs) |\n| image→3D | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fpro\u002Fimage-to-3d` ($0.375, up to 8 views) | `tripo3d\u002Fh3.1\u002Fimage-to-3d` ($0.20) |\n| text→3D | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fpro\u002Ftext-to-3d` | `tripo3d\u002Fh3.1\u002Ftext-to-3d` |\n| retopology | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fsmart-topology` ($0.75) | `tripo3d\u002Ftripo\u002Fremesh` (~75x cheaper) |\n| part split | `fal-ai\u002Fhunyuan-3d\u002Fv3.1\u002Fpart` (FBX only) | `tripo3d\u002Ftripo\u002Fsegment` |\n| text→image | `fal-ai\u002Fnano-banana-pro` ($0.15 flat) | `fal-ai\u002Fflux-2-pro` ($0.03\u002FMP) |\n\nSeedance is ~4x Kling o3's price for the same 5 seconds. It earns that on\nmulti-reference fidelity (up to 50 mixed image\u002Fvideo\u002Faudio refs) and does **not**\nearn it when conditioning on a single still.\n\nModel IDs and prices drift. Verify against fal.ai\u002Fmodels before promising any of\nthem.\n\n## The 3D Branch\n\n```bash\npython mint3d.py --genre \u003Cname> --from-stills 4 --retopo --render\npython mint3d.py --genre \u003Cname> --prompt \"a cracked chrome visor\" --render\n```\n\n- **Generate a clean plate first; do not lift from reference stills.** The\n  endpoint's stated input requirement is simple background, single object,\n  object >50% of frame. Reference reels are the opposite of that — collages,\n  wide shots, several subjects, burnt-in graphics — and they produce sculpted\n  noise. Text → single-object plate → mesh costs ~$0.15 extra and is the\n  difference between a usable mesh and a discarded one.\n- **Multi-view is named per-angle fields, not a list.** `input_image_url` (front,\n  required), then `back_image_url`, `left_image_url`, `right_image_url`,\n  `left_front_image_url`, `right_front_image_url`, `top_image_url`,\n  `bottom_image_url`. There is no `input_image_urls` and no `multi_view` flag —\n  inventing them degrades every mint to single-view while appearing to work. A\n  wrong angle label is worse than omitting the view, because the model trusts it.\n- **Address the GLB by key, not by position.** The response carries a `thumbnail`\n  PNG and a `model_urls` block alongside `model_glb`; taking the first URL works\n  only until the keys reorder, and a preview PNG downloads fine — nothing fails\n  until Blender refuses to open it.\n- **Request PBR maps.** Without them the mesh lights like painted cardboard.\n- **Retopologise** if anyone will edit or rig it. Generated meshes are dense and\n  chaotic.\n- **Render locally to close the loop.** Once a turntable is frames, it is\n  footage, and every downstream stage already handles footage — grade it, cut it,\n  screen it as an element, or upload it as a conditioning reference. Use Blender\n  when a binary is on PATH; keep a dependency-light software rasteriser as the\n  default, because a headless GL context is the single most common thing missing\n  from a container and a renderer that only works on a workstation is not part of\n  a pipeline.\n\n## Verify, Then Believe\n\nEvery other stage claims a result. Check it, and check **distribution shape**,\nnot just moments:\n\n- `background` — share of frame below L\\*10 vs the **pack's** figure. Compare to\n  the reference, **not** to the source clip: a generated source at 68% black is\n  blacker than any reference in a 24–55% band, so \"preserve the source's blacks\"\n  demands the wrong thing and equally excuses a lifted grade.\n- `chroma_mae` — per-zone a\\*\u002Fb\\* error, using the **median** (matching how the\n  pack's targets were measured; a mean here compares a skew-sensitive statistic\n  to a robust one and reports a definition mismatch as an error).\n- `contrast` \u002F `black_point` \u002F `white_point`\n- `banding` — empty L\\* histogram bins *between occupied ones*. Counting total\n  empty bins does not work: a legitimately dark clip has empty highlight bins.\n- `cadence` — detected mean shot length vs the reference's.\n\nWatch the **units trap**: OpenCV changes Lab convention with dtype. On float32,\nL\\* is 0–100 and a\\*\u002Fb\\* are signed; on uint8, L\\* is 0–255 and a\\*\u002Fb\\* are\nbiased +128. Mixing them reports chroma errors in the hundreds.\n\n## Full Chain\n\n```bash\npython pipeline.py --genre \u003Cname> --refs a.mov b.mov \\\n  --brief \"what happens\" --duration 12 \\\n  --base-video existing.mp4 --out out\u002FFINAL.mp4\n```\n\nmint → distill → (mint3d) → apply → forge → verify. Stages 1–3 are cached, so\niterating on briefs never re-measures anything. `--dry-run` stubs every network\ncall: the plan, prompts, track layout and manifest all still get exercised.\n\n## Anti-Patterns\n\n| Don't | Why |\n|---|---|\n| One generation per shot | 21% efficiency, 12 unrelated clips |\n| Put colour in the prompt | Measured not to work; pushes away from the neutral base the LUT wants |\n| Stream-copy the cuts | Keyframe-only boundaries destroy sub-second rhythm |\n| Cut a base video contiguously | Reassembles the original; every cut invisible |\n| Trust compose to overlay | It cannot; it rejects the second track |\n| Send seconds to compose | Silently 1000x too short |\n| Ship on MAE alone | Add the background-share check |\n\n## Borrowed Footage Carries the Capture App's UI\n\nThe single worst defect found in a delivered cut: the finished video shipped\nsomeone else's like button, view counter and comment bubble, because the base\nfootage was a screen recording and nothing cropped them out.\n\n**`content_mask` does not solve this and its bounding box makes it worse.**\nTemporal variance keeps interface chrome, because chrome *animates* - the heart\npulses, the counter ticks - so the mask marks it as moving content. Measured on\nthree references, the mask bbox kept 100% of the width every time while the\ninterface sat plainly in the right-hand margin.\n\nThe separating signal is the temporal **median**, not the variance. Footage\nmoves, so the median of many frames averages into mush with almost no edge\nenergy; chrome sits at fixed coordinates, so its edges survive intact. Sobel\nenergy on the median frame lights up on chrome and goes quiet on content -\nmeasured, a right-hand column read 0.23 against an interior background of 0.03\non one reference and 0.31 against 0.15 on another.\n\nTwo implementation details that cost a cycle each:\n\n- **Trim past the innermost outlier in each outer band, not inward from the\n  edge.** Walking in while the current line is hot stops immediately, because\n  the outermost lines are letterbox - flat black, zero edge energy - and the\n  chrome sits *inside* that at 90-95% of width. The naive version trimmed 1%\n  of frame while the like button stayed in shot.\n- **Scale to cover, not pad,** when portrait source lands in a landscape cut.\n  Padding 9:16 (narrower still after the UI crop) into 16:9 left ~60% of frame\n  as black bars, one shot was nearly an empty rectangle, and it poisoned the\n  background metric because bars are pure black. Covering loses the sides,\n  which is the right trade for centre-framed material.\n\n## Overlay Plates Are Elements, Not Washes\n\nA glow plate is ~4% covered by construction. Composite it at frame size and you\nget a small bright dot parked mid-shot - it reads as a sticker, and it was\nvisible in a delivered cut as an unexplained coloured blob.\n\n- **Tighten each plate to its alpha bounding box first.** That raises coverage\n  from ~4% to 15-35% and hands size control to the caller instead of inheriting\n  whatever fraction of the source frame the element happened to occupy.\n- **Choose wash vs element by coverage.** Diffuse plates (\u003C10% after tightening)\n  work stretched full-frame at low opacity; concentrated ones want to be scaled\n  to 35-70% of frame width and placed.\n- **Vary placement, scale and rotation per shot** from a seeded RNG, so the cut\n  stays reproducible but no two stamped shots share a mark. One plate in one\n  spot every Nth shot reads as a watermark.\n- Resolve element geometry in Python, not in ffmpeg expressions: `pad()` rejects\n  a negative offset and cannot pad below its input size, so an oversized or\n  off-frame element kills the whole filtergraph.\n\n## Downstream Handoff (Resolve \u002F Blender)\n\n**Always ship an editable timeline beside the mp4.** The flattened video is a\nviewing copy and the one thing a colourist cannot work with — every cut is baked\nin and the shots are no longer separable. `forge.py` writes FCPXML 1.9 and a\nCMX3600 EDL referencing the individual graded shot files, so the piece lands as a\ntimeline that can be re-cut and re-graded.\n\nValidate the export, do not assume it: check that asset-clip offsets equal the\nrunning sum of prior durations (no gaps), that every `ref` resolves to a declared\nasset, that every `media-rep src` exists on disk, and that the total matches the\nmp4. A timeline that imports but drifts is worse than one that fails loudly.\n\nTwo things that silently destroy the work:\n\n- **Project frame rate must be set before import.** Resolve locks timeline fps on\n  first timeline creation and conforms the cadence silently. At a sub-second mean\n  shot length that conform is visible.\n- **The delivered shots are already graded.** `look.cube` is a *normalising* LUT\n  for new material and for matching — applying it to the supplied shots\n  double-grades them. Node 1, nothing before it, corrections after.\n\nShip a handoff doc with the measured targets in it (`scripts\u002FHANDOFF-TEMPLATE.md`\nis a filled example): the zone chroma table, the cadence distribution including\n**rhythm variance** — an editor who matches the mean but not the variance\nproduces something that reads completely differently — and an explicit \"do not\"\nlist.\n\nFor Blender, `scripts\u002Fblender_prop.py` derives its lighting from the pack:\nblack world with transparent film (so renders composite with no keying), key plus\nrim (a key alone lets the silhouette die against black), rim colour converted\nfrom the pack's peak-chroma zone via Lab→linear sRGB so the prop picks up the\nsame cast the footage is graded to. **View transform Standard, not AgX\u002FFilmic,\nand no grading in Blender** — AgX applies its own tone curve before the LUT ever\nsees the pixels, and grading twice compounds.\n\n## Bundled Code\n\n`scripts\u002F` in this skill is a working implementation, not pseudocode. It has no\nproject-specific assumptions: point it at any reference videos and it produces a\npack.\n\n```bash\npip install -r scripts\u002Frequirements.txt\nexport FAL_KEY=...            # only needed for the stages that call fal\n```\n\nEvery network call is stubbed under `TASTE_FORGE_DRY_RUN=1` or `--dry-run`, so\nthe plan, prompts, track layout and manifest can be inspected without spending.\n",{"name":5,"metadata":24,"description":7},{"origin":25},"ECC","main","15fa5201935d2b1d172b4e61c2360c908d876d20","plugins\u002Ffrontend\u002Fskills\u002Ftaste-application",0,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP","https:\u002F\u002Fwww.npmjs.com\u002Fpackage\u002Fsfp-hive","https:\u002F\u002Fgithub.com\u002Fsiddu1004","I am a student at Amrita Vishwa Vidyapeetham pursuing my BTECH degree in Electronics and Electrical Engineering (2024-2028).","2026-11-09 16:04:40",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add siddu1004\u002FSFP --skill taste-application",{"tool":41,"command":42},"bunx","bunx skills add siddu1004\u002FSFP --skill taste-application",{"tool":44,"command":45},"pnpm","pnpm dlx skills add siddu1004\u002FSFP --skill taste-application","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002FSKILL.md",true,[49,55,59,63,67,71,75,79,85,89,93,97,101,105,109,113,116,120,124,127,131,135,139,143,147,151,155,159,163,167,171,175,179,183,187,191,195,199,203,207,211,215,219,223,227,231,235,239,243,247,251,255,259,263,267,271,275,279,283,289,293,297,301,305,309,313,317,321,325,329,333,337,341,345,349,353,357,361,365,369],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"scripts\u002F.gitignore","SCRIPT","脚本",25,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002F.gitignore",{"path":56,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":57,"sourceHtmlUrl":58},"scripts\u002Fapply.py",27117,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fapply.py",{"path":60,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":61,"sourceHtmlUrl":62},"scripts\u002Fblender_prop.py",15779,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fblender_prop.py",{"path":64,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Fdistill.py",20283,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fdistill.py",{"path":68,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Ffalapi.py",31178,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ffalapi.py",{"path":72,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":73,"sourceHtmlUrl":74},"scripts\u002Fforge.py",16845,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fforge.py",{"path":76,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":77,"sourceHtmlUrl":78},"scripts\u002FHANDOFF-TEMPLATE.md",8665,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002FHANDOFF-TEMPLATE.md",{"path":80,"resourceType":81,"resourceTypeDesc":82,"sizeBytes":83,"sourceHtmlUrl":84},"scripts\u002FLICENSE","LICENSE","许可证",1071,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002FLICENSE",{"path":86,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":87,"sourceHtmlUrl":88},"scripts\u002Fmint.py",7979,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fmint.py",{"path":90,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":91,"sourceHtmlUrl":92},"scripts\u002Fmint3d.py",11922,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fmint3d.py",{"path":94,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":95,"sourceHtmlUrl":96},"scripts\u002Fpipeline.py",7342,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fpipeline.py",{"path":98,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":99,"sourceHtmlUrl":100},"scripts\u002Fpyproject.toml",617,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fpyproject.toml",{"path":102,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":103,"sourceHtmlUrl":104},"scripts\u002Frequirements-live.txt",242,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Frequirements-live.txt",{"path":106,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":107,"sourceHtmlUrl":108},"scripts\u002Frequirements.txt",66,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Frequirements.txt",{"path":110,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":111,"sourceHtmlUrl":112},"scripts\u002Fresolve_ingest.py",19743,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fresolve_ingest.py",{"path":114,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":29,"sourceHtmlUrl":115},"scripts\u002Ftaste\u002F__init__.py","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002F__init__.py",{"path":117,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":118,"sourceHtmlUrl":119},"scripts\u002Ftaste\u002Fassemble.py",11847,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fassemble.py",{"path":121,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":122,"sourceHtmlUrl":123},"scripts\u002Ftaste\u002Fcadence.py",11414,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fcadence.py",{"path":125,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":69,"sourceHtmlUrl":126},"scripts\u002Ftaste\u002Ffalapi.py","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Ffalapi.py",{"path":128,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":129,"sourceHtmlUrl":130},"scripts\u002Ftaste\u002Fframes.py",11985,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fframes.py",{"path":132,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":133,"sourceHtmlUrl":134},"scripts\u002Ftaste\u002Fgrade.py",33442,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fgrade.py",{"path":136,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":137,"sourceHtmlUrl":138},"scripts\u002Ftaste\u002Fpack.py",5422,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fpack.py",{"path":140,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":141,"sourceHtmlUrl":142},"scripts\u002Ftaste\u002Fplates.py",9983,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fplates.py",{"path":144,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":145,"sourceHtmlUrl":146},"scripts\u002Ftaste\u002Frender3d.py",10461,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Frender3d.py",{"path":148,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":149,"sourceHtmlUrl":150},"scripts\u002Ftaste\u002Fresolve.py",297,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Fresolve.py",{"path":152,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":153,"sourceHtmlUrl":154},"scripts\u002Ftaste\u002Ftimeline.py",20837,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftaste\u002Ftimeline.py",{"path":156,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":157,"sourceHtmlUrl":158},"scripts\u002Ftasteforge\u002F__init__.py",926,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002F__init__.py",{"path":160,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":161,"sourceHtmlUrl":162},"scripts\u002Ftasteforge\u002F__main__.py",172,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002F__main__.py",{"path":164,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":165,"sourceHtmlUrl":166},"scripts\u002Ftasteforge\u002Fapply.py",9679,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fapply.py",{"path":168,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":169,"sourceHtmlUrl":170},"scripts\u002Ftasteforge\u002Fassets.py",10970,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fassets.py",{"path":172,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":173,"sourceHtmlUrl":174},"scripts\u002Ftasteforge\u002Fcli.py",8794,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fcli.py",{"path":176,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":177,"sourceHtmlUrl":178},"scripts\u002Ftasteforge\u002Fcontract.py",25760,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fcontract.py",{"path":180,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":181,"sourceHtmlUrl":182},"scripts\u002Ftasteforge\u002Fdistill.py",5811,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fdistill.py",{"path":184,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":185,"sourceHtmlUrl":186},"scripts\u002Ftasteforge\u002Fexport.py",10073,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fexport.py",{"path":188,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":189,"sourceHtmlUrl":190},"scripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fcadence.json",8019,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fcadence.json",{"path":192,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":193,"sourceHtmlUrl":194},"scripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fflashethereal-cut.edl",8674,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fflashethereal-cut.edl",{"path":196,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":197,"sourceHtmlUrl":198},"scripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fgrade.json",7324,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fgrade.json",{"path":200,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":201,"sourceHtmlUrl":202},"scripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fgrounding.txt",970,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fgrounding.txt",{"path":204,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":205,"sourceHtmlUrl":206},"scripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fpack.json",1377,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fpack.json",{"path":208,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":209,"sourceHtmlUrl":210},"scripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fspec.json",1303,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ffixtures\u002Fflashethereal\u002Fspec.json",{"path":212,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":213,"sourceHtmlUrl":214},"scripts\u002Ftasteforge\u002Fintegration.py",19371,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fintegration.py",{"path":216,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":217,"sourceHtmlUrl":218},"scripts\u002Ftasteforge\u002Finterview.py",3316,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Finterview.py",{"path":220,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":221,"sourceHtmlUrl":222},"scripts\u002Ftasteforge\u002Fmedia\u002F__init__.py",85,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fmedia\u002F__init__.py",{"path":224,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":225,"sourceHtmlUrl":226},"scripts\u002Ftasteforge\u002Fmedia\u002Fcapcut.py",2873,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fmedia\u002Fcapcut.py",{"path":228,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":229,"sourceHtmlUrl":230},"scripts\u002Ftasteforge\u002Fmedia\u002Fcommon.py",1330,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fmedia\u002Fcommon.py",{"path":232,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":233,"sourceHtmlUrl":234},"scripts\u002Ftasteforge\u002Fmedia\u002Fglitch.py",7177,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fmedia\u002Fglitch.py",{"path":236,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":237,"sourceHtmlUrl":238},"scripts\u002Ftasteforge\u002Fmedia\u002Fmanim_geo.py",5167,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fmedia\u002Fmanim_geo.py",{"path":240,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":241,"sourceHtmlUrl":242},"scripts\u002Ftasteforge\u002Fmedia\u002Fstills.py",3196,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fmedia\u002Fstills.py",{"path":244,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":245,"sourceHtmlUrl":246},"scripts\u002Ftasteforge\u002Fpack.py",6775,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fpack.py",{"path":248,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":249,"sourceHtmlUrl":250},"scripts\u002Ftasteforge\u002Fprovenance.py",5827,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fprovenance.py",{"path":252,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":253,"sourceHtmlUrl":254},"scripts\u002Ftasteforge\u002Fproviders.py",2892,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fproviders.py",{"path":256,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":257,"sourceHtmlUrl":258},"scripts\u002Ftasteforge\u002FREADME.md",15666,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002FREADME.md",{"path":260,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":261,"sourceHtmlUrl":262},"scripts\u002Ftasteforge\u002Fresolve.py",12576,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fresolve.py",{"path":264,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":265,"sourceHtmlUrl":266},"scripts\u002Ftasteforge\u002Fschema.py",15315,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fschema.py",{"path":268,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":269,"sourceHtmlUrl":270},"scripts\u002Ftasteforge\u002Ftimeline.py",3579,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Ftimeline.py",{"path":272,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":273,"sourceHtmlUrl":274},"scripts\u002Ftasteforge\u002Fworkflow.py",34614,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Ftasteforge\u002Fworkflow.py",{"path":276,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":277,"sourceHtmlUrl":278},"scripts\u002Fverify.py",10054,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fverify.py",{"path":280,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":281,"sourceHtmlUrl":282},"scripts\u002Fworkflow_graphs.py",6735,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fscripts\u002Fworkflow_graphs.py",{"path":284,"resourceType":285,"resourceTypeDesc":286,"sizeBytes":287,"sourceHtmlUrl":288},"SOURCE.md","REFERENCE","参考文档",2169,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002FSOURCE.md",{"path":290,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":291,"sourceHtmlUrl":292},"tests\u002Ftest_apply.py",10584,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_apply.py",{"path":294,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":295,"sourceHtmlUrl":296},"tests\u002Ftest_assets.py",8375,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_assets.py",{"path":298,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":299,"sourceHtmlUrl":300},"tests\u002Ftest_cli.py",16127,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_cli.py",{"path":302,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":303,"sourceHtmlUrl":304},"tests\u002Ftest_distill.py",3286,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_distill.py",{"path":306,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":307,"sourceHtmlUrl":308},"tests\u002Ftest_integration.py",30206,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_integration.py",{"path":310,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":311,"sourceHtmlUrl":312},"tests\u002Ftest_interview.py",3119,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_interview.py",{"path":314,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":315,"sourceHtmlUrl":316},"tests\u002Ftest_media_utilities.py",16706,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_media_utilities.py",{"path":318,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":319,"sourceHtmlUrl":320},"tests\u002Ftest_multimodal_contract.py",21158,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_multimodal_contract.py",{"path":322,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":323,"sourceHtmlUrl":324},"tests\u002Ftest_multimodal_workflow.py",18035,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_multimodal_workflow.py",{"path":326,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":327,"sourceHtmlUrl":328},"tests\u002Ftest_offline_fixture.py",2688,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_offline_fixture.py",{"path":330,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":331,"sourceHtmlUrl":332},"tests\u002Ftest_pack.py",3481,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_pack.py",{"path":334,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":335,"sourceHtmlUrl":336},"tests\u002Ftest_provenance.py",3194,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_provenance.py",{"path":338,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":339,"sourceHtmlUrl":340},"tests\u002Ftest_providers.py",1773,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_providers.py",{"path":342,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":343,"sourceHtmlUrl":344},"tests\u002Ftest_resolve.py",10281,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_resolve.py",{"path":346,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":347,"sourceHtmlUrl":348},"tests\u002Ftest_schema.py",3982,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_schema.py",{"path":350,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":351,"sourceHtmlUrl":352},"tests\u002Ftest_timeline_export.py",4623,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Ftests\u002Ftest_timeline_export.py",{"path":354,"resourceType":285,"resourceTypeDesc":286,"sizeBytes":355,"sourceHtmlUrl":356},"workflows\u002FREADME.md",4553,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fworkflows\u002FREADME.md",{"path":358,"resourceType":285,"resourceTypeDesc":286,"sizeBytes":359,"sourceHtmlUrl":360},"workflows\u002Ftaste-apply-motion.json",5504,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fworkflows\u002Ftaste-apply-motion.json",{"path":362,"resourceType":285,"resourceTypeDesc":286,"sizeBytes":363,"sourceHtmlUrl":364},"workflows\u002Ftaste-apply.json",5271,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fworkflows\u002Ftaste-apply.json",{"path":366,"resourceType":285,"resourceTypeDesc":286,"sizeBytes":367,"sourceHtmlUrl":368},"workflows\u002Ftaste-distill.json",4690,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fworkflows\u002Ftaste-distill.json",{"path":370,"resourceType":285,"resourceTypeDesc":286,"sizeBytes":371,"sourceHtmlUrl":372},"workflows\u002Ftaste-prop3d.json",2302,"https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Ffrontend\u002Fskills\u002Ftaste-application\u002Fworkflows\u002Ftaste-prop3d.json",[374,381,388,395,402,409],{"skillNo":375,"name":376,"nameZh":6,"description":377,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":378,"categories":379,"keywords":380,"publishedAt":19},"01M7FEEDADZBNBVY19H3XZKCBQ","scientific-schematics","Generates scientific diagram drafts using Nano Banana 2 AI with smart iterative refinement. Uses Gemini 3.7 Flash for quality review. Refines when the review requests improvement, with at most two generations. Specialized in neural network architectures, system diagrams, flowcharts, biological pathways, and complex scientific visualizations.","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Fscience-hardware\u002Fskills\u002Fscientific-schematics\u002FSKILL.md",[],[],{"skillNo":382,"name":383,"nameZh":6,"description":384,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":385,"categories":386,"keywords":387,"publishedAt":19},"01M7FEE0JS04FE58MST48D0MMC","polyhaven-texture-apply","Apply realistic PBR textures from PolyHaven to any Blender object. Supports metals, wood, concrete, fabric, and more. Trigger when asked to texture an object, apply a material from PolyHaven, make something look like metal\u002Fwood\u002Fmarble, or change object surface.","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Fscience-hardware\u002Fskills\u002Fpolyhaven-texture-apply\u002FSKILL.md",[],[],{"skillNo":389,"name":390,"nameZh":6,"description":391,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":392,"categories":393,"keywords":394,"publishedAt":19},"01M7FEDJRWV194TY411WVRS84W","polyhaven-studio-setup","Set up a professional product photography studio in Blender using PolyHaven HDRIs and PBR ground materials. Trigger when asked to create a studio setup, add HDRI lighting, set up product photography, or create a showroom scene.","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Fscience-hardware\u002Fskills\u002Fpolyhaven-studio-setup\u002FSKILL.md",[],[],{"skillNo":396,"name":397,"nameZh":6,"description":398,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":399,"categories":400,"keywords":401,"publishedAt":19},"01M7FED5H2FGS50XGETMGSM0RV","polyhaven-scene-builder","Build complete product scenes using PolyHaven assets — HDRI environment, textured ground\u002Fpedestal, and optional props from the 3D model library. Trigger when asked to build a product scene, create a showcase setup, add a pedestal, or compose a product shot.","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Fscience-hardware\u002Fskills\u002Fpolyhaven-scene-builder\u002FSKILL.md",[],[],{"skillNo":403,"name":404,"nameZh":6,"description":405,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":406,"categories":407,"keywords":408,"publishedAt":19},"01M7FECSNZ2CJA7Z8TPE5R844R","polyhaven-material-swap","Cycle through multiple PolyHaven materials on a product and render each variant. Great for showing color\u002Ffinish options. Trigger when asked to show material variants, swap textures, create color options, show finish comparisons, or render product in different materials.","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Fscience-hardware\u002Fskills\u002Fpolyhaven-material-swap\u002FSKILL.md",[],[],{"skillNo":410,"name":411,"nameZh":6,"description":412,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":413,"categories":414,"keywords":415,"publishedAt":19},"01M7FECBPQ6YWH48PYAMS0C5CE","polyhaven-hdri-showcase","Render the same product across multiple HDRI environments and output a comparison grid. Trigger when asked to compare lighting setups, test different environments, show product in multiple settings, or create an HDRI comparison.","https:\u002F\u002Fgithub.com\u002Fsiddu1004\u002FSFP\u002Fblob\u002Fmain\u002Fplugins\u002Fscience-hardware\u002Fskills\u002Fpolyhaven-hdri-showcase\u002FSKILL.md",[],[]]