跳到正文
cccskills
免费在 GitHub 公开

scan-to-model

Coordinate evidence based reconstruction from scans, photos, plans, and measurements into a reviewable Blender architectural model, with photographic appearance, renders and portable browser delivery when requested.

中文简介正在准备中。当前显示原文说明。

集成工具Blender
查看安装方法

包含的文件(1)

  • SKILL.md7.5 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Use this skill for an end to end scan to model task. Keep source intake, depth inspection, surface measurement, registration, modeling, and visual review as separate evidence records. Preserve original source files, link claims to source IDs and identify every claim as observed, measured, inferred, owner supplied, historical, or unknown. Use existing tools for mechanical work and choose the fastest, lightest capable model for each bounded task.

Read evidence-workflow.md for shared provenance, uncertainty, coverage, and modeling rules. Route archive work to scan-ingest, photograph-first construction inventories to source-observe, supplemental exterior panoramas to street-view-reference, native depth questions to depth-inspect, rigid alignment to scan-register, and Blender edits/review to blender-reconstruct.

  1. Inspect the existing project's source ledger, model, accepted transforms, owner corrections and unresolved coverage. Preserve work already done; do not restart the reconstruction or inherit an unverified transform. Record the requested finish and outputs, plus existing permission to interpolate. Distinguish a geometry study, a rendered approximation and independent metric acceptance. A geometry-only request does not acquire a rendering or browser requirement.
  2. State a bounded next increment: one room with its ceilings/connections, or one named missing connection. Name the sources, reviewable outcome, acceptance check and finite stopping condition. Default to one active geometry increment, including its review, correction and delivery, plus one independent source/records lane. If candidates are already waiting, select one to finish and keep the rest queued; do not batch-review the backlog or start another build merely because workers are available. Expand only with named independent acceptance owners and demonstrated capacity. Proceed with previously authorized work; ask only for missing information that changes the result.
  3. Ingest new archives, examine contact sheets spanning the entire capture and sharp source frames, then update the coverage atlas. For a fresh whole-house build, identify the observed rooms, levels, stairs, outdoor areas and their connections before choosing a footprint. Sample more densely at transitions and unresolved regions; a few evenly spaced photographs cannot establish complete coverage. Keep photo examination distinct from depth support.
  4. Use depth-inspect to establish camera axes, gravity, tracking segments and level heights before combining RGB-D points. Measure physically identified floors, walls, ceilings and openings within each region; use these measurements to drive the layout. A point-cloud bounding box is not a floor plan. Validate room/level joins with opaque shared features and independent views before accepting them as measured. Keep authorized tentative joins distinct from accepted registration. Interpolate remaining gaps around supported dimensions and topology; do not substitute a generic house template for recoverable evidence.
  5. Build and inspect a saved Blender candidate. Check its major physical relationships against source views before polished drawings or presentation. For a rendered house, take one representative region through shape, materials and a corrected final-size render using rendered-reconstruction.md before repeating its construction helpers or finishes across the house. Preserve unrelated architecture and contents. Deliver supported changes and authorized tentative choices with their status visible in the model; report exactly what remains provisional.
  6. Complete required delivery checks, then update the existing handoff and ledger together once per increment. Reuse checks bound to unchanged inputs; do not create additional review or receipt layers merely to certify prior bookkeeping. Continue through the authorized rooms and requested finish; completing one bounded increment does not finish a whole-house request. Before claiming completion, reconcile the source coverage with the saved model, inspect the final renders and exported membership, and read the latest delivery/browser results. Report omissions and failures even when a command exits successfully.

For requested appearance, still tours, packed native files or web fly-throughs, use delivery.md and the shared scripts/delivery.py producer. Keep geometry, appearance, view review and packaging status separate. When interpolation is already authorized, fill inconclusive visible geometry or finishes with a reasoned tentative choice, recording its basis; do not repeatedly ask for the same permission. Preserve missing-evidence and metric limits. A complete-looking model does not promote inferred construction to observed fact. Respect explicit exclusions such as no movies, and retain only the source dependencies actually used when the owner excludes unused intake directories.

Keep delivery records compact according to delivery.md. An owner's request to retain working evidence overrides those defaults. Keep private project evidence outside the plugin package.

A photograph-first inventory can proceed while the base geometry or registration is under review. Keep source observations and tentative feature identities separate from replaceable model placements; see source-observations.md.

Use project-records.md for source, room/connection, observation, registration and model records. Use the package scripts through formats.md. These are local tools; keep private project evidence outside the plugin package. If delegating, give mechanical intake/index tasks a precise contract and independently verify counts, timestamps and marked pixels before using their outputs as geometry evidence.

For outcome-first dispatch, stopping rules and reusable worker/reviewer briefs, use work-packets.md. For the goal shape and the current project-supplied completion example, use goals.md. Keep the parent milestone separate from the investigation outcome, bind immutable inputs and fresh outputs, and state the next gate and escalation conditions. Reopen a completed investigation only for identified new evidence, a corrected invalid method or a materially different question that could change the decision; cosmetic retry variations are not new evidence. Improve a tool or reusable instruction when it removes a demonstrated obstacle to the current deliverable, then apply and verify that improvement on the actual task. Publishing authorization and private project data remain project-owned.

Current finishes lead current geometry; historical open wall material supplies services only where spatially reconciled. Keep unsupported connections unresolved unless the owner authorized tentative interpolation; label any such interpolation in the model. Ask for a specific missing physical fact only when it is necessary for the requested result and cannot be recovered or handled within that authorization. Do not add approval gates or depend on Superpowers.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Build and review evidence tagged Blender architectural candidates, preserve unrelated scene content, and produce requested photographic materials, stills and portable model deliveries.

中文简介正在准备中。当前显示原文说明。

obra/scan-to-model72026年9月27日 更新

Inspect native RGB, depth, confidence, calibration, and poses to make finite, source traceable reference points and surface measurements.

中文简介正在准备中。当前显示原文说明。

obra/scan-to-model72026年9月27日 更新

Preserve and inventory scan archives while retaining capture variants, hashes, calibration, and frame provenance.

中文简介正在准备中。当前显示原文说明。

obra/scan-to-model72026年9月27日 更新

Fit and independently validate rigid scan alignment from named, source traceable correspondences.

中文简介正在准备中。当前显示原文说明。

obra/scan-to-model72026年9月27日 更新

Use when cataloging or annotating visible construction in photographs or scan RGB, especially before registration or while the base model may be replaced.

中文简介正在准备中。当前显示原文说明。

obra/scan-to-model72026年9月27日 更新

Collect and verify Google Street View exterior references for a building reconstruction, preserving panorama identity, imagery dates, attribution and viewpoint separately from calibrated scan evidence.

中文简介正在准备中。当前显示原文说明。

obra/scan-to-model72026年9月27日 更新

查看 obra 的全部技能

报告此技能的问题