Compress accumulated LLM context locally via a port 8790 proxy.
npx skills add https://github.com/yoonhwan/ai-feature-pack --skill headroom-yoonhwan
Please help me install this Agent Skill. Skill: headroom Source: https://github.com/yoonhwan/ai-feature-pack/tree/main/feature-pack/headroom Command: npx skills add https://github.com/yoonhwan/ai-feature-pack --skill headroom-yoonhwan
Long AI coding sessions accumulate massive context that gets resent every turn, burning through cache and rate limits. headroom compresses that context locally before it reaches the LLM, cutting re-prefill costs and extending productive work time before hitting hard limits.
/headroom on so it never activates without explicit user consent.Use the headroom skill to enable context compression for your current project by running /headroom on and then launch Claude Code through the claude-hr wrapper to start saving tokens on long sessions.
High-intent search queries and answers about installing and using this skill.
You can reduce token costs in long AI coding sessions by compressing accumulated context locally before sending it to the LLM, cutting down re-prefill spend and rate-limit consumption. headroom applies this compression automatically.
Context compression preserves prefix cache hits by using deterministic output alignment to shrink the middle context band while keeping the prefix intact. This maintains cache efficiency while reducing overall token transmission size.
To enable context compression for a specific project, run the `/headroom on` command to opt in via the per-project registry, then launch Claude Code through the `claude-hr` wrapper to start routing through the local proxy and saving tokens.
If the local LLM proxy crashes during a coding session, the fail-open wrapper logic automatically falls back to a direct connection, preventing session-wide outages and ensuring your AI coding work continues without interruption.
Yes, context compression applies to multi-agent swarms and repeated document processing where context re-prefill dominates spend. By shrinking accumulated context before transmission, it extends productive work time before hitting hard rate limits.
arakitakashiSet up and launch the homework-coach-robo local environment with Docker Compose or direct execution.
tharunramagiriInstall and configure the rtk token-compression proxy for Bash tool interactions.
kangarookingManage token budgets and context windows for long-running AI conversations.
cwinvestmentsMonitor Headroom proxy health and report compression statistics for Claude Code sessions.