菜单
Preserve prompt-prefix stability so the provider's KV cache keeps hitting, cutting cost and latency on long sessions. Use when structuring system prompts, memory files, or tool definitions, when a long session is getting expensive, or when deciding where to put changing vs stable content.
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
Compress heavy context before it costs tokens. Use when the user attaches or asks to read a PDF, scanned document, screenshot/large image, or a verbose log / build output / CI output — convert PDFs to Markdown, compress oversized images, and digest noisy logs so the same content costs a fraction of the tokens.
Keep the main conversation cheap by offloading heavy sub-tasks to isolated subagent contexts. Use when a sub-task (research sweep, large-file investigation, log triage, broad search) will generate a lot of intermediate output whose conclusion is small — branch it to a subagent so the verbose work never enters the main thread, and merge back only a compact digest.
Find code by symbol name instead of reading whole files. Use when you need to locate a function, class, or method, understand where something is defined, or navigate an unfamiliar repo — query the jusTokenMax code index to get file:line + signature in a few tokens, then read only that range.
Cut the tokens the agent WRITES, not just what it reads. Activate when the user asks for terse/concise/brief output, says responses are too long, wants to save output tokens or cost, or runs /justokenmax:terse — then answer in a compressed style that preserves all technical substance while dropping filler.
| name | cache-align |
| description | Preserve prompt-prefix stability so the provider's KV cache keeps hitting, cutting cost and latency on long sessions. Use when structuring system prompts, memory files, or tool definitions, when a long session is getting expensive, or when deciding where to put changing vs stable content. |
Providers cache the KV state of a stable prompt prefix; a cache hit is far cheaper and faster than recomputing it. Any edit near the START of the context invalidates everything after it. This skill is the discipline that keeps the prefix stable so the cache keeps paying off.
If a long session is getting expensive, check for prefix churn first: a memory file being rewritten each turn, a timestamp in the system prompt, tool definitions being regenerated. Fixing prefix stability often saves more than any single compression.
