All tags
Posts tagged with "Developer Workflow"
Claude Code Skill Safety: From 'Please Stop' to 'You Can't Move'
38 Skills, three layers of defense, one hard lesson: natural language instructions are not a safety mechanism. How I systematically hardened 12 unprotected destructive Skills with PreToolUse Hooks, Skill splitting, and disallowed-tools.
Claude Code Skill 安全性：從「拜託你停下來」到「你根本動不了」
38 個 Skills、三層防護、一個血淚教訓：自然語言指令不是安全機制。本文記錄我如何用 PreToolUse Hook、Skill 拆分和 disallowed-tools 系統性地修補 12 個毫無 checkpoint 的破壞性 Skills。
Safety Gates in Claude Code Skills: From Auditing 35 Skills to a Three-Layer Protection Model
I assumed writing 'Use AskUserQuestion' in a Skill was a hard constraint. After auditing 35 Skills, reading the official docs, and digging through GitHub Issues, I found out: the model uses the same mechanism to decide whether to obey your CHECKPOINT and whether to invoke your tool. There's only one gate that's truly 100%.
Claude Code Skill 的安全閘門：從 35 個 Skills 的審計到三層防護模型
我以為在 Skill 裡寫 Use AskUserQuestion 就是 hard constraint。審計完 35 個 Skills、查完官方文檔和 GitHub Issues 之後發現，模型用同一套機制決定要不要理你的 CHECKPOINT 和要不要調用你的 tool。真正 100% 的閘門只有一個。
Git as an External Brain for Claude Code: Beyond MEMORY.md
MEMORY.md isn't the end of the road for AI Agent memory. When project scale exceeds what a context window can hold, Git becomes the truly scalable external memory. This post breaks down the three layers of memory, Git's role among them, and which practices have research backing vs. which are just my own experiments.
Git 作為 Claude Code 的外接大腦：超越 MEMORY.md 的記憶架構
MEMORY.md 不是 AI Agent 記憶的終點。當專案規模超過 context window 能承載的範圍，Git 才是真正能無限擴展的外接記憶體。這篇拆解記憶的三個層次、Git 在其中的角色、以及哪些做法有研究支撐、哪些只是我自己的實驗。