Free prompt optimizes repeated agent workflows, cutting 90%+ tokens
vivekhaldar · x · 2026-10-05
A developer recommends tokenshrinker, a free tool: paste its optimization prompt into Codex, Claude Code, or any agent that can read past sessions, and it will identify one repeated workflow, build an optimized version, and show measured before-and-after results.
In tests across 15 historical scenarios, gross tokens dropped 94–96%, elapsed time fell 87–92%, and AI calls dropped up to 10x, with useful output quality unchanged. No install or account required; optimized versions land in a separate folder and nothing is applied without approval.
More from coding & agent
- Amazon (2B) and Cloudflare (9B/27B) ship open-source decision models rivaling Jev — mark_k · 2026-10-05
- Autonomous agents only pay off where value justifies engineer escalation, dev argues — Pavel_Asparagus · 2026-10-05
- Users report agents like Codex feel noticeably slower than three months ago — koltregaskes · 2026-10-05
- Building After Effects motion graphics engines with Opus 5.5 — notiansans · 2026-10-05
- Local inference in practice: mining repos, AI chat logs and media libraries with a tiny model — natesiggard · 2026-10-05
- repOx packs 3k-file repos into LLM prompts in ~14ms, stripping 30k-token lockfile waste — Wvdy_CC · 2026-10-05