Skip to content

v4.8.0

Choose a tag to compare

View all tags
@KimYx0207 KimYx0207 released this 10 Jul 14:48
v4.8.0
753ff43

Solved Problem

The v4.7 playbook forced an external Skill-search ritual and could label a successful native Agent dispatch as fallback when an optional Skill was not installed. That made healthy Codex and Claude Code execution look degraded.

Changes

  • Stop capability discovery on the first qualified local Agent, Skill, Tool, Command, MCP provider, or capability-index match.
  • Run external find-skills discovery only after a real local capability gap is proven.
  • Use Codex's current top-level spawn_agent(task_name, message, fork_turns) contract without legacy typed or namespaced fields.
  • Keep Claude Code on its native Agent / Task contract, require the Agent prompt, and use Team surfaces only when the host exposes them.
  • Add synchronized Claude Code and Codex runtime packages, release metadata, regression tests, and repository ignore rules for secrets and local runtime state.

Verification

  • node --test tests/runtime-contracts.test.mjs — 6/6 passed.
  • bash -n scripts/install.sh — passed.
  • Live Claude Code 2.1.196 regression — one Skill call and one Agent call with required prompt, no retry.
  • Codex runtime evidence — three successful native top-level Agent spawns.
Assets 2
Loading