v1.1.0 — every proven Claude Code token optimization, in one curl command

[sravan27]

Hi everyone — v1.1.0 is out. If you are hitting the 5-hour rate window on Claude Pro/Max, this is for you.

What it does

One curl command installs 12 proven token optimizations:

1. Response shaping (CLAUDE.md) — 40-65% fewer output tokens
2. Output style terse — deeper contract than CLAUDE.md, enforced every turn
3. .claudeignore — blocks node_modules, dist, .next, target, 60+ noise dirs + lock files
4. Secret exclusion — .env, *.pem, credentials.json, SSH keys
5. Repo map + stack hints — Next.js, Python, Rust, Go, Flutter detected automatically
6. Thinking budget cap — MAX_THINKING_TOKENS=8000 (default is 32K+)
7. Early compaction — at 80% instead of 95%
8. statusLine — live model · branch · context-os ✓ on every prompt
9. Slash commands — /compact, /context, /ship
10. Haiku explorer subagent — 15x cheaper exploration in isolated context
11. Output compression hooks — 27-70% on test/build output (optional binary)
12. Session memory hooks — decisions survive compaction (optional binary)

See what you would save (no install)

curl -fsSL https://raw.githubusercontent.com/sravan27/context-os/main/setup.sh | bash -s -- --measure

Scans your repo, prints a conservative per-session estimate. No writes.

Install

cd your-project
curl -fsSL https://raw.githubusercontent.com/sravan27/context-os/main/setup.sh | bash

Zero dependencies. Idempotent. Reversible (--uninstall preserves your own CLAUDE.md content).

Real benchmark

scripts/benchmark.sh runs the same task against a repo before and after install, parses claude --print --output-format json, reports measured delta. On a 2-file fixture: 32% token reduction, 18.8% cost reduction. Bigger repos benefit more.

Feedback welcome

If you know a technique that measurably reduces Claude Code token consumption and it is not in the 12-row table, open an issue with benchmark numbers. The goal is completeness.

Release: https://github.com/sravan27/context-os/releases/tag/v1.1.0