context-pack
Keep development token use low by refreshing and relying on local context packs.
When & Why to Use This Skill
The Context Pack skill is a specialized developer tool designed to optimize AI agent performance and reduce API costs by managing local context efficiently. It enables the agent to generate, refresh, and rely on a centralized local context file (context_pack.md), ensuring high-fidelity project awareness without excessive token consumption. This is particularly effective for maintaining focus during complex software engineering tasks and long-running development sessions.
Use Cases
- Large-scale Code Refactoring: Consolidate project knowledge into a local pack before major structural changes to keep the agent focused on relevant logic while minimizing token overhead.
- Post-Feature Synchronization: Refresh the local context after completing a feature chunk to provide the agent with an updated 'source of truth' for subsequent development steps.
- Token Usage Optimization: Significantly lower development costs in long sessions by directing the agent to prioritize local context files over re-reading the entire codebase.
- Session Compaction Preparation: Use the skill to summarize current progress and open questions into decision notes before compacting the conversation history, preventing loss of critical project state.
| name | context-pack |
|---|---|
| description | Keep development token use low by refreshing and relying on local context packs. |
| allowed-tools | Bash, Read, Grep, Glob, Write |
Context Pack Skill
When to use
- Before large refactors
- After finishing a feature chunk
- Before /compact
What to do
- Run /ctx-refresh (Gemini with cooldown)
- Read context/local/context_pack.md
- Base next steps only on that file + files you touch
- If uncertain, write an OPEN QUESTION into context/local/decision_notes.md