Profile
Plugin that intercepts shell commands via pretoolcall and rewrites output through RTK, compressing terminal output before it reaches the LLM context window. 60-90% token reduction on shell commands, 96.6% efficiency across 11M+ tokens processed. Zero config — auto-loads on gateway boot. Real benchmarks: cargo test 90-99%, git log --stat 87%, ls -la 78%.
Signals
Listed in the awesome-hermes-agent README
Sources: 2 / Surfaces: 1
What the upstream surface says
Short excerpt only, so you can decide whether to click out.
[](https://pypi.org/project/rtk-hermes/) [](https://opensource.org/licenses/MIT)
Transparently rewrites shell commands executed via Hermes's terminal tool to their RTK equivalents, achieving 60-90% LLM token savings.
Agent runs: terminal(command="cargo test --nocapture") → Plugin intercepts pretoolcall hook → Calls rtk rewrite "cargo test --nocapture" → Mutates args["command"] = "rtk cargo test --nocapture" → Agent executes the rewritten command → Filtered output reaches LLM (~90% fewer tokens)
- rtk-rewrite
- RTK binary not found → plugin disabled silently
- rtk rewrite times out (>2s) → command passes through unchanged
- rtk rewrite crashes → command passes through unchanged
- No RTK equivalent → command passes through unchanged