菜单
Use right after fixing any bug or discovering a reusable pattern. If it could bite again on different work, auto-add it to AGENTS.md or .adn/lessons.md in the same response, without asking.
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。
Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.
Use after every turn. For every piece of data that entered context (user prompts, tool results, file contents, sub-agent reports, errors, code read), evaluate whether it has reusable value and auto-document it without asking.
Use when producing any text a person other than you will read (README, CHANGELOG, GitHub issue or PR text, hand-off notes, .adn docs). Expand every abbreviation on first use and tailor content to the reader. Banned placeholders TBD/TBC/???.
Use when bringing AI documentation infrastructure (AGENTS.md rules, .adn docs, shared skills, hooks, agents, settings) from a richer source repo into a leaner or solo-managed target repo, and source-specific content must be stripped and strict enforcement relaxed. Also use to re-sync after the source repo's infra changes.
Use when answering the user in chat. Default to 1-3 sentence, plain-language replies with no preamble and no trailing summary. The user has repeatedly asked for short, simple answers.
| name | auto-gotcha |
| description | Use right after fixing any bug or discovering a reusable pattern. If it could bite again on different work, auto-add it to AGENTS.md or .adn/lessons.md in the same response, without asking. |
When any fix or pattern discovered during work is reusable for the future, AUTOMATICALLY record it immediately after the fix. Do NOT wait for the user to ask.
AGENTS.md.adn/lessons.md (dated entry)Why: The reflex is to fix the issue and move on without recording the reusable part, which means the same class of mistake can recur. SSH Lite already states this as a rule ("same mistake twice = broken process"); this skill is the mechanical habit that makes it happen in the same response, not "next time".
How to apply: After every fix, ask: "Could this bite us again on a different change?" If yes, add it to the right place right away, in the same response as the fix. No confirmation needed. Before saving, check whether AGENTS.md / .adn/lessons.md already covers it and update rather than duplicate.
Counter-example (the failure mode this prevents): a fix is made, the root cause is understood, but the reusable insight is only kept in the chat and never written down. A later change hits the same trap and the work is redone from scratch. The cost of one line of documentation now is far less than re-debugging later.