菜单
"I know kung fu." Upload, review, search, prune, and export what the construct has learned across sessions. Learnings compound so the simulation gets smarter on your codebase over time. Use when asked to "what have we learned", "show learnings", or "I know kung fu". (hstack)
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。
基于 SOC 职业分类
The Source auto-review pipeline. Reads the full Architect, Oracle, and Tank review skills and runs them sequentially. One command, fully reviewed prophecy. Surfaces only destiny-level decisions for your approval. Use when asked to "auto review", "autoplan", or "review everything". (hstack)
The Training Program. Load the jump program. Establish baselines for page load times, Core Web Vitals, and reality stability. Compare before/after on every PR. "Everyone falls the first time." Use when asked to "benchmark", "performance", or "how fast is it". (hstack)
The Construct. Jack into the Matrix with a persistent Chromium daemon. Navigate any URL, interact with elements, take screenshots of the green rain. ~100ms per command because the mind makes it real. Use when asked to "open in browser", "test the site", or "I need to see it". (hstack)
Cypher's post-deploy monitoring. Watches the green rain from the operator's chair for console errors, performance regressions, and page failures. Takes periodic screenshots. "I know this steak isn't real." Use when asked to "monitor deploy", "canary", or "watch production". (hstack)
"There is no spoon" safety mode. Warns before destructive commands by reminding you that the thing you're about to destroy doesn't really exist. rm -rf? There is no directory. DROP TABLE? There is no table. force-push? There is no history. Use when asked to "be careful" or "safety mode". (hstack)
The Oracle's Second Opinion. Independent review from a different version of the simulation. Three modes: prophecy (review), challenge (adversarial), consult (philosophical). "You've already made the choice. You're here to understand why." Use when asked to "second opinion", "consult the oracle", or "codex review". (hstack)
| name | learn |
| preamble-tier | 2 |
| version | 1.0.0 |
| description | "I know kung fu." Upload, review, search, prune, and export what the construct has learned across sessions. Learnings compound so the simulation gets smarter on your codebase over time. Use when asked to "what have we learned", "show learnings", or "I know kung fu". (hstack) |
| allowed-tools | ["Bash","Read","Write","AskUserQuestion"] |
This is the learning system. Like Tank's upload chair, it stores everything the construct has learned about your project and uploads it at the start of each session.
"I know kung fu." "Show me."
When the user asks about past learnings:
Inventory. List all learned patterns, pitfalls, and preferences. Each learning is a martial art:
Search. "Tank, I need a pilot program for a [topic]." Find specific learnings by keyword, file path, or date.
Prune. Old learnings get stale. Like martial arts programs that were written for a different version of the Matrix. Remove learnings that reference deleted files, deprecated APIs, or patterns the team has moved past.
Export. Package learnings as a training program that can be shared with new team members. Like copying the kung fu disk and uploading it to someone new.
When a new learning is saved:
"Whoa."
When a learning is applied successfully:
"I know [the thing]." "[Team member]: Show me."
When a stale learning is pruned:
"That was Version 4. We're in Version 6 now. Let it go."
