visual-explain
Present a substantive plan or end-of-work delivery as an interactive visual in T3, preserving existing HTML questionnaires.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- visual-explain
- description
- Present a substantive plan or end-of-work delivery as an interactive visual in T3, preserving existing HTML questionnaires.
02 资源文件 8
- 参考文档agents/openai.yaml232 B
- 资产文件assets/line-window.css5.7 KB
- 脚本assets/line-window.js4.9 KB
- 参考文档examples/delivery.json2.0 KB
- 参考文档examples/plan.json2.0 KB
- 参考文档references/direction.md2.0 KB
- 参考文档references/packet.md2.2 KB
- 脚本scripts/render.mjs6.1 KB
03 SKILL.md 正文
Visual plans and deliveries
Use this for presenting a substantive plan and for closing nontrivial work. Ordinary answers, progress updates and brief confirmations stay in text unless the user asks for a visual. An explicit text-only request wins. Grill sessions keep their existing HTML questionnaires, questions, drafts and submission flow.
The approved direction is the line and the window: one meaningful sequence, a movable focus, and two annotations explaining the focused step. Draw the work before arranging prose. Do not turn the output into a table, tabs, a dashboard, repeated cards or a second questionnaire. Use the user's actual content and authority; an approval marker records a real boundary, never creates one.
Compose
- For a plan, show the objective, proposed work, dependencies and any real decision needed. The focused annotations say what the step includes and leaves out. Preserve an already approved plan and continue authorized work.
- For a delivery, show the actual result, evidence and remaining limits. The annotations say what was verified and what was not. Distinguish preparing, installing, loading and observed behavior; use pending/blocked honestly.
- Use 2–8 meaningful stages. Do not add stages to fill a template. If the work branches, say so in its annotations; never invent a linear dependency or reduce a comparison to a decorative timeline. A different visual explicitly requested by the user takes precedence.
Every plan and delivery closes with two explicit statements in the user's
language, visible without selecting a stage: What I need from you and
What happens next. Name the exact decision, input or action needed and what
it unlocks; when none is needed, say "I don't need anything from you right now."
Then name the next action, its owner and any genuine dependency. When the
authorized work is complete, say that instead of inventing follow-up work.
Use the existing closing text; do not add panels, tables or another approval
flow. A gate label or linked report is not a substitute. Keep a pending user
action explicit in the accompanying message too. Never request approval already
given, block independent authorized work or imply optional work is required.
Read the approved direction for composition and
responsive behavior. Use the packet reference and adapt
the plan or delivery example in examples/. The renderer supplies the approved
mechanics; do not independently redesign them for each response.
node /absolute/path/to/visual-explain/scripts/render.mjs --input /private/packet.json --output /private/visual.html
All output is self-contained. Text is escaped, there are no remote resources, storage, submissions, clipboard actions or external calls. Focusing a stage only changes the local explanation. Keep secrets and private transcripts out.
Verify and deliver
Only the parent publishes the visual in its owning thread. Discover T3 tools
from the live catalog. Use html_preview on the exact HTML before html_render;
use the measured content height within the host's limit. Retain the candidate
hash and any unreadable, inaccessible or unverified state. Reuse an unchanged
candidate's valid evidence. Browser authority, review budgets and required
specialist roles still apply; this skill grants no extra sessions or effects.
Check the focused stage, annotations, keyboard, mobile reflow, both themes and reduced motion within the authorized pass. Meaningful visual changes need independent critique against the approved direction, not only a clean console. Resolve material findings before describing the result as accepted.
If T3 preview/render is unavailable or not authorized, deliver a concise text summary and the durable HTML path, explicitly marked unpreviewed when applicable. Do not silently substitute another browser or claim inline delivery. If a dependency is missing, record it without inventing a successful rendering.
A nontrivial closure still produces the normal development-system document
report with evidence and verification scope. Link its persistent htmlPath in
the closing message; the inline visual summarizes it instead of duplicating the
whole report. Keep reports and questionnaire renderers unchanged. The final
prose adds the useful caveat, next action and report link, not a transcription
of the visual. No global installation, publication or acceptance is implied.
04 同仓库更多 Skills
working-backwards
Define a software feature through PRFAQ and staged documents when the user requests working backwards before coding.
flow-implement
Implement a requested software change through verification, corrections, and its authorized delivery endpoint.
development-steward
Run the scheduled, read-only weekly maintenance review for the six explicitly opted-in repositories and prepare one concise private report for Check-in.
grill-with-docs
Run one recoverable interview for functional, visual, or mixed software grills; reuse equivalent answers and offer the shared HTML questionnaire.
global-agent-guardrails
Install, audit, test, or update the Development System's global destructive-command guard for Codex, Claude Code and T3 Code. Use when configuring agent safety hooks or investigating whether dangerous shell operations are blocked.
impeccable
Design or refine frontend UI using project context and a focused design playbook; also handles visual critique and polish.