Hand-crafted Claude Code skills for higher-quality agent results.
A curated set of context-engineering skills: when to compress, when to summarize, when to spawn a subagent, how to checkpoint long sessions. Built and battle-tested by NeoLab. Compatible with OpenCode, Cursor, Antigravity, Gemini CLI.
claude mcp add context-engineering-kit-skill -- git clone https://github.com/NeoLabHQ/context-engineering-kit ~/.claude/skills/context-engineering-kit
一行命令搞定。用 claude mcp list 驗證,claude mcp remove 移除。
使用場景
實戰用法: Context Engineering Kit
Keep a multi-hour coding session from blowing the context window
👤 Engineers on long agent sessions⏱ ~15 minintermediate
何時使用: You're 3 hours into a coding loop and the model is starting to drop context.
前置條件
Server/skill installed and authenticated — See repo README
步驟
Checkpoint
Use the context-engineering skill's 'checkpoint' to summarize what we've done into a tight handoff doc.✓ 已複製
→ A handoff doc Claude can re-read instead of full history
Resume
Compact the conversation and resume from the handoff doc.✓ 已複製
→ Session continues with fresh context
結果: Sessions that don't degrade as they grow.
注意事項
Checkpointing too often loses fine-grained context. The kit suggests every ~30 turns or before any /compact. — Checkpointing too often loses fine-grained context. The kit suggests every ~30 turns or before any /compact.
成本與限制
運行它的成本
API 配額
See provider docs for rate limits
每次呼叫 Token 數
Varies by tool
費用
See repo README for pricing details
提示
Cache tool results and avoid repeated identical calls.
安全
權限、密鑰、影響範圍
憑證儲存: Use environment variables; never commit secrets
資料出站: Tool calls go to the provider's API as documented
故障排查
常見錯誤與修復
checkpoint doc too long
Tune the skill's max-length; default is ~1500 tokens. Anything bigger and the doc itself eats context.
驗證: wc -w on the checkpoint
subagent doesn't have enough context
The skill's spawn template requires a 'mission' block — fill it explicitly, don't let it default.