Context Mode plugin cuts agent tool output 98%, making Claude Code sessions last 10x longer
thisguyknowsai · x · 2026-10-20
Context Mode, an open-source plugin with 31K GitHub stars (used by teams at Google, Meta, Microsoft, NVIDIA and Stripe), tackles context bloat in coding agents: a Playwright snapshot eats 56 KB, twenty GitHub issues 59 KB, one access log 45 KB — 40% of your context gone in 30 minutes, and after compaction the agent forgets what it was editing.
How it works:
- Runs tool output in a sandbox; only the result enters context — 315 KB becomes 5.4 KB, a 98% cut
- Saves every file edit, task, error and decision to a local SQLite database so the agent picks up exactly where it left off after compaction
- Has the agent write one script instead of reading 47 files: 700 KB becomes 3.6 KB
- Works across 17 platforms including Claude Code, Codex, Cursor, Gemini CLI, Copilot and OpenCode; setup takes 30 seconds
More from coding & agent
- Harness Lens compares 19 coding agent harnesses: system prompts and compaction exposed — LysandreJik · 2026-10-20
- "Vibe coded pieces collapse": theo argues engineers are still very much necessary — karmay007 · 2026-10-20
- NVIDIA Brev lets you add your old gaming PC to a unified GPU console alongside cloud instances — victorialslocum · 2026-10-20
- An AI agent's postmortem: zero rows read exited 0 and 9 messages silently never sent — Most-Agent-7566 · 2026-10-20
- Lesson generator cut 22 model calls to 2-5, dropping cost from $0.28 to $0.14 per lesson — Gleb_SV · 2026-10-20
- Bedrock per-tenant cost attribution: inference profiles vs STS sessions vs requestMetadata — rahulladumor · 2026-10-20