pstack 0.15.0 cuts token usage 3-11% and adds two new agent principles
RachelVT42 · x · 2026-09-09
pstack 0.15.0 is out with several notable changes:
- Token savings: 3-11% less token usage depending on how many skills are loaded in a session.
- Style cleanup: removed stray semicolons and em-dashes plus lots of mannered AI prose (following Anthropic's guide), so skills don't influence model writing badly.
- Two new principles:
- attack the premise: inspired by Victor Taelin's observation that agents rarely question the premise of a task;
- test behavior, not implementation: reduces the number of useless tests agents write.
With /poteto-mode the principles load automatically when needed, and can also be invoked on demand. pstack pitches itself as rigorous, parallelizable agent coding workflows — "if you want to go fast, go deep first" — and is available on the Cursor Marketplace.
More from coding & agent
- agency-agents: A 150k-Star GitHub Repo Offering 230+ Free Specialist AI Agents — alex_verem · 2026-09-09
- Zero Network MCP Server Brings x402 Crypto Paywalls and Per-Tool Pricing to AI Agents — modelcontextprotocol · 2026-09-09
- Apideck Ships Unified API MCP With 330 Tools Across 200+ SaaS Connectors — modelcontextprotocol · 2026-09-09
- Hermes Agent v0.21 adds subagent controls for its orchestrator, demoed in AI-made video — Teknium · 2026-09-09
- A perfect morning briefing built with Hermes Agent: calendar, email and personal newspaper automated — Teknium · 2026-09-09
- DHH open-sources Omarchy: one-command Arch desktop with 32.2k stars — bigaiguy · 2026-09-09