npx agentspec-cli install 9e48736a-0e77-4d75-92d4-d33158d4178c --framework ClaudeCode→ .claude/skills/hance/SKILL.md
# /hance A single entry point that routes to one of six subcommands: `setup`, `run`, `try`, `refine`, `batch`, `ui`. ## Routing Pick the subcommand from the user's request. When in doubt, ask. | User intent | Subcommand | Doc | |---|---|---| | "install hance" / "set up hance" | `setup` | `subcommands/setup.md` | | "apply <preset> to <file>" / "grade with <preset>" | `run` | `subcommands/run.md` | | "make this look like X" / "match this reference" / "show me some looks for this" | `try` | `subcommands/try.md` | | "make this one look better" / "tune/refine this grade" / "more grain, warmer, etc." | `refine` | `subcommands/refine.md` | | "apply X to all of these" / batch a folder | `batch` | `subcommands/batch.md` | | "open the editor" / "open the UI" | `ui` | `subcommands/ui.md` | ## Runner selection (every subcommand) Pick one runner per invocation, in order: 1. If `command -v bun` succeeds → use `bunx @orva-studio/hance` (preferred — fast cold-start, cached after first fetch). 2. Else if `command -v node` succeeds → use `npx @orva-studio/hance`. 3. Else → inform the user that a compiled binary is available for download from GitHub releases, but do **not** install it for them. Runner priority: Bun (`bunx`), then Node (`npx`), then compiled binary as a last resort. Never install the binary for the user — just let them know it exists if they have neither Bun nor Node. Every subcommand must work on a fresh machine; `setup` only verifies runtime + ffmpeg and shows examples. The compiled binary expects a `hance-gpu` sidecar alongside it; a bare `./hance` checkout without it fails with `ENOENT … hance-gpu`. This is why `bunx`/`npx` are preferred — they always work. When developing in this repo, run the source directly: `bun run packages/cli/src/cli.ts …`. ## Hard rules - Never invent flags. Only use names listed in `hance --help` or `hance preset --help`. - Never run a full video render to explore looks — use `hance preview` (still frame) for `try`. - Windows is unsupported. If `uname -s` reports Windows, stop and tell the user. ## References - `references/preset-index.md` — `presets/index.json` schema and rebuild contract. - `references/compare-page.md` — `/compare` route query params and Edit hand-off. - `references/grading.md` — how to judge and dial in a film look: the render→read→adjust loop, what makes a grade read as film, and common artifacts. Read before `refine`.
Built on the Agent Skills standard
media-use
Agent Media OS for a HyperFrames project. Resolve BGM, SFX, image, icon, brand logo, voice, color grade, or LUT into a frozen local file or paste-ready block + ledger record (one verb, `resolve`); generate via TTS / music / image models when the catalog misses; produce voiceover, transcription, captions, and background removal through one shared audio engine; operate on media (cut / reframe / transform); and reuse assets across projects. Also use for vague feedback that real footage looks dark, flat, boring
by AgentHub Crawler
AI Agent Guidelines for Montage AI
This file provides essential context for AI coding agents working on this codebase. It should be read in conjunction with `docs/llm-agents.md`.
by AgentHub Crawler
hyperframes
Mandatory entry point: read this first for any request to make, create, edit, animate, or render a video, animation, or motion graphic, including a promo, explainer, captioned clip, title card, overlay, slideshow or interactive deck, Remotion port, or any HyperFrames HTML composition. Also use it to inspect, diagnose, validate, preview, publish, or batch-render an existing HyperFrames project. Inputs may be a website URL, GitHub PR, Figma design or URL, text or brief, existing footage, or music. It resumes
Sign in to leave a comment.
by AgentHub Crawler