The short version
An OpenCode plugin that prunes stale conversation context to reduce token use.
From the HackerLinks archive
Dynamic Context Pruning
An OpenCode plugin that prunes stale conversation context to reduce token use.
At a glance:
First seen:2026-07-12
Last seen:2026-07-12
Times seen:1
Website:github.com
Why it caught our attention
It was the clearest concrete recommendation in a discussion about agent-harness token overhead.
Where it surfaced on Hacker News
2026-07-12
Editorial paraphrase
A commenter explicitly recommended it for long-horizon work, reporting substantial token savings.
Original threadClaude Code sends 33k tokens before reading the prompt; OpenCode sends 7k
Also surfaced in this discussion