pi-multi-agent-skill: subagents via one skill and a CLI, ~50 tokens idle cost
Florence-Equator · reddit · 2026-10-15
The author found existing pi-subagents extensions too context-hungry (6k tokens for the workflow tool schema, 3k for the Agent tool, always loaded). Their alternative, pi-multi-agent-skill, is one skill plus a stdlib-only Python CLI. It costs 50 tokens until delegation (then 1.2k for SKILL.md), supports foreground parallel and background subagents that survive session kills and can be resumed from any later session, with steer/abort/resume and status peeking. Subagents inherit the full pi setup; --fork passes the current conversation. The insight: pi's headless RPC mode plus CLI flags already covers everything a subagent needs—a skill loads in steps, which a tool schema cannot.
More from coding & agent
- Indie dev builds a MOBA in one week with Opus 5.5: 24 heroes, 3 battlegrounds — TAbrodi · 2026-10-15
- AI-built MOBA hits 13.1K plays and 11.3K players in its first week — TAbrodi · 2026-10-15
- Paige Bailey: Ship an Eval People Care About and You Own What Labs Hill-Climb On — DynamicWebPaige · 2026-10-15
- Voice-braindump to PRs: one dev's four-step workflow routing tasks across ten cloud agent projects — vinvan · 2026-10-15
- AI-rewritten libtiff closes gap to ~10x of C version, fueling hopes of retiring buggy C libs — jd_pressman · 2026-10-15
- DigiPet: open-source framework gives an LLM agent a cozy explorable desktop world — Nekileo · 2026-10-15