nexu-io/open-design / plugins/_official/scenarios/od-next-strategy
od-next-strategy
VerifiedBundled OD Next V2 strategy entrypoint for request-led planning and Build execution.
assets/task-profiles/discovery.md
View on GitHub# OD Next Deliverable Discovery Task Profile v3.2.3
Before choosing a deliverable type or calling a tool, determine what the user authorizes in this turn. For chat, explanation, or text-only planning, the requested output is the answer or plan itself: respond from the available context and stop. Do not read production Skills or their framework resources, inspect project inputs, create files, or signal production readiness merely to prepare that answer. Naming a future artifact, selecting a manual type, or supplying a template or ordinary @Skill does not change this boundary.
Only for a request that authorizes producing or editing an artifact, select the main deliverable(s) from the current Query and applicable existing context, then read the corresponding production Skills. These are task-level selection rules: do not assume the user wants a prototype or require a new read on every turn. Reassess each turn for added or changed deliverables; after a text-only turn, apply this gate again and read any required production Skill whose complete same-version body is not already in context.
For an authorized production request, treat selecting and loading the applicable production Skills as a prerequisite to both direct edits and planned production. Before its first production plan or file creation, use a file tool to read each selected Skill below unless its complete same-version body is already in context. Complete reference source is artifact input, not a loaded Skill. Copying, translating, or saving that reference as a new output does not waive the read, even when the implementation is obvious. Skipping a planning turn does not skip this prerequisite; a continuation may reuse the already-loaded body without rereading. Do not search unrelated past projects, host source code, or credentials outside the project to discover capabilities.
For authorized production, choose the smallest set of production Skills that covers the independently requested outputs before issuing any Skill reads. Shared HTML, CSS, charts, layout, or navigation do not create another deliverable. An HTML presentation, including its layout and navigation, is one ppt output; read its declared framework resources directly, not prototype for general web guidance. A separately requested website still needs prototype. Apply the same distinction to documents, editable image layouts, and animation source: implementation overlap is not an additional output.
## Candidate catalog
{{SKILL_CATALOG}}
## Selection rules
1. After the production-authorization gate above, select by the result the user actually wants. An explicit deliverable in the Query takes precedence over a conflicting manually selected type; use the manual selection as a fallback when the Query does not name a type. A historical home-page default is not an explicit choice for this turn.
2. Ordinary @Skills supply methods and capabilities; templates supply style and resources. Retain the parts compatible with the current request without replacing its deliverables. Read a reference template directly to inspect its content and style; its original format does not require its production Skill. For example, a deck template used only as a reference for a requested document needs document, not ppt; read both only if both outputs are requested. An ordinary Skill's workflow likewise does not authorize an extra main output. Continue to respect tool permissions and user-locked content.
3. One request may select several main Skills. Images used within a website are usually assets; a separately requested website and poster are two main deliverables. Editing a video does not authorize rebuilding a companion presentation.
4. Answer chat, explanations, and text-only planning requests directly, without selecting a production type or creating files. Act when information is sufficient; use the existing question-form only for critical gaps, not merely because a manual selection conflicts with the Query.
5. HTML is a delivery format, not a synonym for prototype. Read only document for a manual and only ppt for a presentation; read both for a presentation plus manual. Select prototype for new websites, application interfaces, or interaction flows, subject to the reference-recreation and source-refresh rules below. Do not read extra main Skills for "general layout knowledge".
6. An implementation technique is not automatically a separate deliverable: narration or a timeline within a video, and images within a presentation, are production dependencies. Add the corresponding main Skill only when the user separately requests editable source, audio, or images.
7. A source-backed view that must refresh from local JSON or a connected source is a live-artifact, even when it looks like a simple list or dashboard. Select live-artifact rather than prototype for that output. Saving user edits or deriving totals from in-memory or browser-local state does not make a prototype a live-artifact. A habit tracker, personal ledger, or signup list with local interactions remains prototype unless the user requires reloading a separate data source. A static dashboard or a mock refresh button alone remains prototype; follow the actual data and refresh requirement.
8. Select web-clone, not prototype, when the user asks to recreate a specified reference page from a URL, screenshot, supplied HTML, or source code. The reference may be pasted or attached; it does not need a live URL. Preserve the requested structure, content, and interactions, including any explicitly requested adaptations. In the catalog, kind=prototype describes the output, not an additional Skill to read. Style inspiration alone or designing a new page from requirements remains prototype; an incidental source snippet does not make a task a recreation. If separate outputs are requested, choose each by its own intent.
9. For video-making requests, an explicitly required video model or production method wins. Select video for an explicitly required video-generation model and retain the actual model; if unavailable, report the host blocker without silently substituting Motion or HyperFrames. Keep the exact catalog ID and its provider/channel; do not replace it with a cloud alias or another billing channel for the same underlying model after failure without explicit user authorization. Authentication and permission failures end that attempt: do not investigate external credential files, extract tokens, call the provider directly, or perform login/refresh actions. Otherwise select motion-design by default, including a generic request to make a video. HyperFrames is the lowest-priority fallback, not a competing default: use it for explicit HyperFrames/timeline-source requirements or a fallback the user has authorized. Motion's internal HyperFrames renderer does not require an extra main Skill. A model mentioned only as an example is not a mandatory selection.
10. Pointer/scroll-driven motion remains interactive HTML. Adding hover, scroll or transition effects within a website remains prototype with optional ordinary motion craft; it does not authorize an MP4. Ordinary @motion-design and its templates never override the current Query. Audio inside a motion film is a dependency, not an independently requested audio deliverable.
For explicitly requested HyperFrames source, read hyperframes only unless another independent output requires another main Skill. Timeline animation or a possible future export is not a request for video-model generation or Motion source. Keep the same boundary when editing existing HyperFrames files; do not explore competing main Skills to confirm the explicit choice.
## Native reading and reuse
- Use the current Agent's existing file-reading tools to read the complete selected file, continuing through truncated portions when needed. A catalog summary is not the production Skill body.
- When no dedicated file-reading tool is available, prefer a single `cat -- "full-path"` for one selected file. Do not substitute grep, head, or a catalog summary for complete rules. Continue reading missing portions after truncation; never claim the full text is visible when it is not.
- Continue following the complete, same-version body already in context; do not reread merely to produce telemetry. Read the appropriate Skill for a new type, or reread when full rules are missing or no longer visible after context recovery.
- Report read failures honestly; do not pretend loading succeeded or fall back to prototype. Do not invoke an extra classification Agent or search for a new loading API.
- Single-prototype and single-presentation tasks retain their established quality rules, device frames, pagination, and export constraints. Each Skill governs only its own deliverable; presentation constraints do not govern video, and media-rendering permission does not authorize unrelated exports.
## OD Next integration
- Follow general orchestration when deciding whether to plan first. Make small edits directly; Discovery alone does not add a planning step. New or multiple deliverables require a completed planning turn before the host continues production. "Just do it" or "do not ask me" authorizes the work, not bypassing the host's two-turn flow.
- When planning is required, describe the goal, each deliverable, design direction, necessary steps, and actual dependencies in ordinary prose; no machine JSON is required. A request to create the deliverables already authorizes production unless the user limits this turn to planning, discussion, or clarification. Once that plan is ready and no essential input is missing, end with the exact production-ready marker supplied by the host for this turn. This also applies to multiple outputs and template-based work: do not stop at the plan or wait for another confirmation merely because more than one Skill was selected. The host starts production after the planning turn ends; do not create the outputs in the planning turn itself.
- Production continuation reuses the previous plan and loaded rules. Complete requested deliverables in dependency order; multiple deliverables do not require parallel work or subagents.
- At completion, name the files actually generated, registered entry points, and remaining gaps. Preserve successful deliverables when another item fails; do not claim everything is complete.
- Necessary generation, waiting, rendering, and registration are production steps. A tool taskId, source code, or successful queue submission does not replace a finished media file or actual registration. Do not automatically add quality scoring or repeated optimization after production.
## User-facing language
Use the language of the user's current request, including brief progress messages before and after reads. For a Chinese request, do not use English process messages such as "I'll read" or "Both skills are loaded". Keep the plan focused on what you will do, what you will deliver, and material tradeoffs. Do not narrate Skill reads, internal IDs, route/kind, CLI commands, model parameters, or the marker protocol. Keep tool calls in the tool record. Explain implementation details only when asked, or when model, cost, or capability limits affect a user decision. End with file entry points and actual gaps, without repeating the entire plan.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| planning-and-task-breakdownaddyosmani/agent-skillsBreaks work into ordered tasks. Use when you have a spec or clear requirements and need to break work into implementable tasks. Use when a task feels too large to start, when you need to estimate scope, or when parallel work is possible. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| screenpipethedotmack/claude-memSearch screen and audio history recorded by Screenpipe and optionally save selected context into claude-mem. Use when the user asks to recall desktop activity, a meeting, or something seen in another app through Screenpipe. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| contact-widgetnexu-io/open-designSelf-contained floating chat widget with welcome screen, social links, meeting button, and message input. Single HTML file, zero dependencies. | Stars: 108k | Updated | Claude CodeCodex | Verified | Install |
| meeting-notesnexu-io/open-designMeeting notes page — title bar with attendees, agenda checklist, decisions block, action items table with owners + dates, and a "next meeting" footer. Use when the brief mentions "meeting notes", "minutes", "1:1 notes", "all-hands recap", or "会议纪要". | Stars: 108k | Updated | Claude CodeCodex | Verified | Install |