Release 2.2.0: GPT-6 routing and proactive Luna fleet #3

2026-09-23T02:45:32Z

kakan commented

2026-09-23 02:45:32 +00:00

Summary

Release 2.2.0 on top of 2.1.0: replace the previous four-tier recommendations with GPT-6 Luna, Sol and Astra. Sol takes over Terra's implementation and investigation workload; Terra is excluded from recommendations and automatic fallbacks while explicit user pins remain respected. Preserve proactive Astra consultation, rolling fleet scheduling, decision-record validation and progress-based expert rounds from 2.1.0.

The optional global prompt explicitly authorizes proactive parallel work for ordinary engineering tasks. The router dispatches bounded Luna assignments early, refills actual available capacity as results return, and keeps shared decisions and final acceptance with the parent. Add GPT-6 async-tool, steering and context-recovery guidance with explicit API-versus-host limitations and dated pricing sources.

Validation

Repository resources, skill-creator validation and git diff checks pass.
Unit tests: 36 of 37 pass on Windows. The symlink fixture fails with WinError 1314 before exercising the installer; the same failure reproduces on the original baseline. No tests were skipped or weakened. Evidence self-test and decision-record example validation pass.
Review covers routing, standing authorization, ownership, runtime capability claims and static acceptance scenarios. No claim of GPT-6 child runtime or performance benchmarking.

Notes

Does not install the skill, replace the user's active prompt or change host configuration. New model routes remain conditional on the actual child-tool schema.

## Summary Release 2.2.0 on top of 2.1.0: replace the previous four-tier recommendations with GPT-6 Luna, Sol and Astra. Sol takes over Terra's implementation and investigation workload; Terra is excluded from recommendations and automatic fallbacks while explicit user pins remain respected. Preserve proactive Astra consultation, rolling fleet scheduling, decision-record validation and progress-based expert rounds from 2.1.0. The optional global prompt explicitly authorizes proactive parallel work for ordinary engineering tasks. The router dispatches bounded Luna assignments early, refills actual available capacity as results return, and keeps shared decisions and final acceptance with the parent. Add GPT-6 async-tool, steering and context-recovery guidance with explicit API-versus-host limitations and dated pricing sources. ## Validation - Repository resources, skill-creator validation and git diff checks pass. - Unit tests: 36 of 37 pass on Windows. The symlink fixture fails with WinError 1314 before exercising the installer; the same failure reproduces on the original baseline. No tests were skipped or weakened. Evidence self-test and decision-record example validation pass. - Review covers routing, standing authorization, ownership, runtime capability claims and static acceptance scenarios. No claim of GPT-6 child runtime or performance benchmarking. ## Notes Does not install the skill, replace the user's active prompt or change host configuration. New model routes remain conditional on the actual child-tool schema.

kakan added 2 commits 2026-09-23 02:45:32 +00:00

Redesign GPT-6 routing and proactive Luna fleet workflow 3a9d09eafc

Integrate 2.1 workflow and prepare release 2.2.0 14c43d0a60

kakan merged commit 9474064daf into main

2026-09-23 02:45:57 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: kakan/codex-subagent-router#3