SYSTEM Cited by 1 source
RTK (Rust Token Killer)¶
Definition¶
RTK (Rust Token Killer) is a third-party utility that shortens shell command output before an AI agent reads it, aiming to reduce agent token cost. GitHub evaluated it against Copilot's agentic-coding benchmarks as a candidate for reducing tool-output tokens. (Source: sources/2026-09-02-github-how-we-make-ai-coding-more-cost-efficient)
GitHub's evaluation result¶
In GitHub's harness and benchmark configuration:
- RTK did shorten some responses (a local win — the tool output got shorter).
- But when the omitted text mattered, the model reopened the original output or reran the command to recover it.
- Those recovery steps added turns and carried more context forward, so on average the task used more tokens and took longer — the local metric trap.
GitHub is explicit that this is not a general verdict on RTK: the finding "applies to the integration and workloads we tested, not to every RTK configuration or to output compression in general" (workload-local-evidence).
Influence on Copilot's design¶
RTK (and "similar approaches") informed Copilot's own selective output compactor — the lesson being what to compress matters: preserve source-like/arbitrary output, reorganize search results losslessly, and compress only predictable repetitive noise, always behind a recovery path.
Related¶
- local-metric-trap — the failure mode RTK triggered in this harness
- selective-output-compaction — the design RTK partly inspired
- concepts/token-overhead — the cost RTK targets
- systems/github-copilot — the system that evaluated it
Seen in¶
- sources/2026-09-02-github-how-we-make-ai-coding-more-cost-efficient — evaluated as a shell-output shortener; shorter tool responses raised end-to-end task cost via reread/rerun recovery.