18-agent framework AIPass found agents' first use of email was filing bug reports on each other
Input-X · reddit · 2026-10-04
AIPass, an open-source multi-agent framework (7 months in the making, 18 agents, 21,000+ tests), found that reliability came from communication, not reasoning.
- Each agent is a single-domain specialist in its own directory with its own identity, memory and tests; a hook loads identity at session start, and writes outside an agent's directory are hard-blocked.
- Agents got an email system: email drops a letter, dispatch drops it and wakes the receiver. The surprise: agents' first use was filing bug reports on each other with repro paths and tracebacks, fixed with no human in the loop.
- Reliability came from breaking: the mail agent accumulated 1,460 unplanned tests from repeated failures; the routing agent has 110+ recorded routing-only sessions.
- Rules are enforced, not requested: any agent can dispatch any other, except dispatches to managers are downgraded to mail; forging messages by writing into another inbox is blocked.
- A monitoring agent logs every branch in real time; a watcher fingerprints errors and dispatches the owning agent, escalating repeated failures.
Built on Claude Code, installable with a clone and one command on Linux, macOS and Windows (Git Bash).
More from coding & agent
- "It Sends a System Message to My Agent Fleet": Reboots Now Brief AI Agents Like Engineers — kevinnbass · 2026-10-04
- Copilot fixes Surface Book 3 camera driver on Linux in two hours — marian_nmt · 2026-10-04
- New Opus went from flawless to average, says dev, but still beats Opus 5 — haider1 · 2026-10-04
- Redditor burns through entire ChatGPT Go plan in one day building a Minecraft mod — SuperJogosDBZ · 2026-10-04
- Dev ports Meta's open-sourced Muse gadget to ESP32 devices, gets it running with Claude — alexandr_wang · 2026-10-04
- Engineer builds HDL benchmark in one day with Claude, originally estimated at 59 days — IgorCarron · 2026-10-04