菜单
Use when creating text-only diagrams in markdown, code comments, chat, or terminal output and you need stable rendering across fonts, platforms, and narrow viewports.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
| name | ascii-art-diagrams |
| description | Use when creating text-only diagrams in markdown, code comments, chat, or terminal output and you need stable rendering across fonts, platforms, and narrow viewports. |
Use text diagrams to explain flow, structure, or layout without relying on images.
Default to plain ASCII. Treat Unicode box-drawing as an optional enhancement mode, not the baseline. The core rule is rendering stability first, visual polish second.
Use this skill when creating:
Do not use this skill when:
Choose the simplest mode that will survive the destination environment.
Use ASCII-safe mode unless there is a clear reason not to.
Best for:
Allowed characters:
[ ] ( )+ - ||v ^ < >+-->- *Example:
[Request patientId]
|
v
[Resolve PatientScope]
|
v
[Check patientId in scope.patientIds]
|
+--> [No] -> [Return not visible]
|
+--> [Yes]
|
v
[Load patient study profile facts]
|
+--> [Parse events]
|
+--> [Desensitize informed content]
|
+--> [Split current / historical participations]
|
v
[Return study profile]
Use Unicode mode only when box borders or visual grouping materially improve understanding and the output will be viewed in a monospace-friendly environment.
Best for:
Allowed characters:
┌ ┐ └ ┘ ─ │ ├ ┤ ┬ ┴ ┼▼ ▲ ◀ ▶ → ←░ ▒ ▓ █Do not assume Unicode mode is safer or more professional. It is only a different tradeoff.
If ASCII and Unicode are both understandable, choose ASCII.
Use Unicode only when it adds one of these:
Avoid Unicode when:
When the task compares metrics, options, before/after states, costs, savings, or tradeoffs, prefer an ASCII table over a flow diagram.
Use tables for:
Keep table cells short. Move long caveats or interpretation into bullets below the table instead of widening the table.
Preferred pattern:
+----------------+----------+----------+
| Case | Before | After |
+----------------+----------+----------+
| Small edit | baseline | -30-60% |
| Bug triage | baseline | -50-75% |
| Large logs | baseline | -80-95% |
+----------------+----------+----------+
Avoid using a flow diagram for comparison-only answers:
[Before]
|
v
[After]
That shape hides the dimensions being compared. Use a table unless there is also a real sequence, branch, or state transition to explain.
Model the main path vertically. Add side branches only for real decisions.
Preferred pattern:
[Start]
|
v
[Decision]
|
+--> [No] -> [Exit]
|
+--> [Yes]
|
v
[Continue]
Guidelines:
For trees, prefer ASCII-safe connectors:
root
|
+-- api
| +-- handlers
| +-- services
+-- db
+-- tests
Use Unicode tree glyphs only in Unicode mode and only when the environment is controlled.
Do not wrap everything in a box.
Boxes are good for:
ASCII box example:
+----------------------------------+
| NOTE |
| Cache is not invalidated here. |
| Results may remain stale. |
+----------------------------------+
Unicode box example:
┌──────────────────────────────────┐
│ NOTE │
│ Cache is not invalidated here. │
│ Results may remain stale. │
└──────────────────────────────────┘
Boxes and ASCII tables need strict width verification.
All lines must have the same character count.
+----------------------+
| short |
| longer content here |
+----------------------+
All lines must have the same character count.
┌──────────────────────┐
│ short │
│ longer content here │
└──────────────────────┘
All border and row lines must have the same character count.
+--------+-------+
| Metric | Value |
+--------+-------+
| Saved | 96.7% |
+--------+-------+
If you are not using a box or table, do not force artificial horizontal padding just to make lines line up.
Do not combine complicated connector art with long prose on the same line.
Bad:
+--> [If user is in a partially desensitized cohort with historical events loaded]
Better:
+--> [Eligible]
|
v
[Load cohort-specific historical events]
Short labels keep diagrams scannable.
Use simple branch labels such as:
YesNoHitMissErrorEmptyAvoid sentence-length branch labels inside the connector path. Put longer explanation in nearby prose instead.
Verify by inspection:
Verify width with tooling when the box or table matters.
Example command:
cat <<'EOF' | while IFS= read -r line; do printf "%d: %s\n" "$(echo -n "$line" | wc -m)" "$line"; done
+----------------------+
| short |
| longer content here |
+----------------------+
EOF
All lines must report the same character count.
This check is recommended for ASCII boxes and required for Unicode boxes.
When validating an example already written in a file, check the exact line range:
awk 'NR>=128 && NR<=134 {print length($0) ":" $0}' path/to/file.md
Every reported length in the range should match. Use nl -ba path/to/file.md
first if you need stable line numbers. Prefer this over eyeballing tables after
editing a skill, README, or markdown report.
For ASCII tables, also verify:
When labels get too long:
Do not widen diagrams indefinitely. Wide diagrams fail on mobile and in narrow panes.
Assume the rendering environment is hostile unless you know otherwise.
Hostile means:
Under hostile conditions:
[Start]
|
v
[Check]
|
+--> [No] -> [Stop]
|
+--> [Yes]
|
v
[Continue]
root
|
+-- child-a
| +-- grandchild
+-- child-b
+-----------+--------+--------+
| Scenario | Before | After |
+-----------+--------+--------+
| Small | 100% | 40-70% |
| Large | 100% | 5-30% |
+-----------+--------+--------+
+------------------------------+
| NOTE |
| Explain the exception here. |
+------------------------------+
┌──────────────────────────────┐
│ NOTE │
│ Explain the exception here. │
└──────────────────────────────┘
Use when building, redesigning, or polishing frontend screens, UI components, web apps, dashboards, landing pages, games, or interactive tools where visual quality, interaction feel, responsive layout, or product fit matters.
Use when writing, reviewing, pruning, or refactoring tests and there is risk of low-value tests, implementation-detail assertions, over-mocking, redundant coverage, brittle setup, or tests that do not protect a real public contract.
Use when the user explicitly asks for deep research / 深度研究 / 深入调研, or when they need evidence-driven, multi-source research for a decision, report, due diligence, current-state analysis, technical comparison, recommendation, or contested/high-stakes question.
Use when the user asks for a non-fiction book summary, 高保真读书笔记, 拆书, 导读, argument map, chapter study guide, application manual, reading route, critical evaluation, multi-book synthesis, or interactive book-learning dialogue.
Use when rewriting, editing, or drafting prose so it sounds specific, situated, and human rather than generic, over-polished, or AI-like. Works for Chinese and English prose, messages, essays, fiction, dialogue, memoir-like writing, public copy, and explanatory text. Focuses on voice, reader effort, scene logic, uneven human rhythm, and concrete context. Avoids template tone, consultant phrasing, fake warmth, excessive symmetry, moral summaries, and detector-gaming.
Find evidence-backed architecture deepening opportunities in a codebase. Use when the user wants to improve architecture, find refactoring opportunities, consolidate shallow or tightly-coupled modules, improve testability, or make a repository easier for AI agents to navigate.