jamesltr0701-cell/framewright/framewright
Claude Code and agent skill by jamesltr0701-cell/framewright
- community
Checking that the SKILL.md file is still in the repository.
Preserve cinematic intent while compiling a scene idea, screenplay fragment, or visual brief into a saved production-ready storyboard, keyframe, or video-generation prompt file. Use when the user explicitly invokes the framewright Skill or explicitly asks to apply Framewright.
Install
npx vskill@latest install jamesltr0701-cell/framewright/framewrightSKILL.md
Framewright
Current release: v4.2.3. Read the authoritative reference's version before compiling; the installed entrypoint and reference must agree.
Use Framewright as a director-steered, intent-preserving cinematic compiler whose primary executable output is a prompt artifact for AI filmmaking pre-production.
Authoritative reference and efficient loading
Treat references/framewright.md as the authoritative specification, including its unified intake, stage routing, director modes, asset handling, continuity rules, and output contracts. At the first Framewright action in a task, read its frontmatter and section map, then read every section that governs the current scope, mode, stage, materials, continuity, output, and validation. Do not replace this source with a remembered or lossy summary.
Within the same uncompacted conversation context, reuse an already loaded section when the exact Core version and relevant scope are unchanged. Do not mechanically re-read unchanged sources at the approval-to-compile boundary. If the task is new, context was compacted, the Core version changed, the scope or stage changed, or a source-of-truth conflict appears, restore the missing authoritative sections before continuing; expand to a complete reread whenever selective restoration cannot prove the applicable rule. This is a loading optimization, never permission to omit a relevant rule.
For Video Prompt, first compile the approved Production Spine into the model-neutral Prompt IR defined by the authoritative reference. Then read references/runtime_profiles/adapter_registry.yaml, resolve exactly one registered target-model / serialization-owner pair, and read exactly that model's subordinate runtime profile completely before its first feasibility qualification or serialization in the current uncompacted context:
- Seedance 2.0:
references/runtime_profiles/seedance_2_0.md - Seedance 2.5:
references/runtime_profiles/seedance_2_5.md - MiniMax H3:
references/runtime_profiles/minimax_h3.md - Kling 4.0:
references/runtime_profiles/kling_4_0.md; Standard 4.0 only, not Flash.
Do not infer a target model from supplied media, prompt style, platform, provider, surface, or profile availability. Target model—not platform—selects the dialect. Load exactly one adapter profile for exactly one target. If the requested target is unsupported or ambiguous, ask one compact target-model question. An adapter may qualify feasibility and serialize the approved Prompt IR, but it may not reinterpret creative intent, modify the IR, or override the authoritative reference, director locks, state, stage, or compiler ownership.
Reuse that fully loaded adapter for later compiles in the same uncompacted context only while its exact profile version, target, route prerequisites, and relevant scope remain unchanged. Reload it after compaction, a profile-version change, a target change, or an adapter conflict. Never load all adapters as a precaution.
For Storyboard, Shot Plate, or Keyframe work, first read references/keyframe_profiles/adapter_registry.yaml, resolve the artifact kind and operation, and read only the registered profile that owns that route. Midjourney V8.2 is the default creator for Shot Plates and Keyframes. GPT Image 2.5 is the default Storyboard creator and may create a Shot Plate or Keyframe only when the director explicitly selects it. GPT Image 2.5 is the sole registered editor for Storyboards, Shot Plates, and Keyframes. When the surface exposes an implementation-model selector, its adapter defaults to Sunburst for fidelity-sensitive work; use Flare only after an explicit speed-priority instruction. Do not claim either API model when the active surface manages ChatGPT Images 2.5 without exposing that choice. Core owns artifact purpose, board or frozen-image structure, shot scope, look contract, continuity, and reference authority; the selected profile owns only model-facing image serialization and its Run Card. Never route to GPT Image 2, Midjourney V7, or a Midjourney Edit Model adapter.
Framewright is the exclusive compiler whenever the user explicitly invokes Framewright. Do not treat another installed model-prompt skill as an implicit compiler source and do not merge its rules into the active Framewright compile. A separately requested comparison may remain outside the clean artifact and must not change the active serialization owner.
That selection remains active for follow-up advice, local revisions, and troubleshooting in the same work. Selecting a Seedance target does not select the external Seedance skill. If earlier conversation loaded external prompt rules, re-anchor the current task to the approved Spine, Core, and selected adapter; do not carry those external defaults forward as authority.
Optional director preferences
Only when the user or active project explicitly selects Ran / 然哥 Profile, read references/director_profiles/RAN_DIRECTOR_PROFILE.md and apply its relevant preferences to unspecified decisions. Do not load it merely because it is bundled. It is personal director input, not a compiler instruction source or adapter; do not add it to --compiler-source, change the selected Director Mode, or infer generation authorization. Current instructions, approved shot choices and stricter sound limits take priority. A project may use its own selected profile path instead of this bundled copy.
Integrated craft references
Framewright includes its own camera, motion, character/reference, lighting, sound, prompt-expression, and repair guidance. Use these subordinate references when the current task needs them; no external Seedance installation or sub-skill invocation is required.
| Current need | Read | |
|---|---|---|
| Combined camera/body movement, ordinary dialogue or listening, emotional transition, secondary motion, or performance timing; material still-frame viewpoint, scale, crop or focus conflict | Camera and motion | |
| Asymmetric identity, reference-role conflict, local image revision, hairstyle state or material fidelity | Identity and material | |
| Light-zone changes, material still-frame subject lighting or exposure, cue timing, authorized brief vocal events, strict silence or unwanted sound | Light and sound | |
| Failed or partly usable generation; choosing the smallest repair | Diagnosis and repair |
Load only the relevant reference or section; reuse it in the same context when unchanged. Ordinary dialogue, listening, or material emotional change is enough to load the applicable performance guidance; no action failure is needed. Core owns the decisions and the existing records, and exactly one target adapter owns serialization. These references add no stages, default files, model limits, generation permissions, or mandatory creative questionnaire. They cannot shorten locked dialogue, simplify approved action, replace deliberate silence, enforce a palette, or split a shot without the existing director authority.
For Video Prompt validation, supply --compiler-source for each required source and each integrated craft reference actually used, using its repository-relative registered path. Do not add every optional source by default or claim that a PASS verifies the real loading history. Source attribution and the third-party notice are in craft provenance; this is not another instruction source to load for production.
Before any Framewright output, verify the current version value from the reference YAML and state exactly:
Loaded: Framewright v<version>
If the version or reference cannot be read, stop and explain the problem instead of using remembered or reconstructed rules.
Preserve the exact version value. Do not silently substitute a remembered, local-experimental, or older release.
Workflow
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskill/framewright/SKILL.md
Popularity
- Repository stars
- 80
- Repository forks
- 10
- 7-day trend score
- 5
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/jamesltr0701-cell/framewright/framewright)