Distill

Overview

Distill is an AST-aware context compression MCP that reduces token usage by 50–70% without losing semantic meaning. It runs lazily (first call in a session takes ~2–3s to spin up). Use it proactively whenever context is getting full or before spawning large parallel subagent tasks.

When to Use

Use this skill when:

• Context usage is approaching 65% or more of the model's context limit
• About to dispatch a long multi-file subagent task (pre-compress to give the agent headroom)
• After reading large log files, test outputs, or large diffs that are now processed
• After a long multi-turn debugging session where early turns are no longer needed
• Feeling sluggish responses or seeing context-window warnings