v0.1.0 — Initial public release
First public release of claude-codex-dispatch — a Claude Code skill that turns OpenAI Codex CLI into a structured dispatch target with role contracts, write-scope policy, and a full audit trail.
What's in this release
- Four dispatch modes:
worker(bounded implementation),verifier,reviewer,synthesizer(the latter three are read-only by contract). - Task packet format with
WRITE SCOPE,NON-GOALS, andVERIFICATIONfields — the orchestrator commits to scope upfront, before Codex runs. - Write-scope enforcement via
git statussnapshots before and after the run; any out-of-scope mutation in worker mode (or any mutation in read-only modes) flipspolicy_violation: trueinpolicy.json. - Full run artifacts at
.codex-dispatch/runs/<id>/:task.md,prompt.md,events.jsonl,result.json,result.md,pre-status.txt/post-status.txt,pre-diff-stat.txt/post-diff-stat.txt,policy.json. - Cross-platform launchers (POSIX
bin/codex-dispatch+ Windowsbin/codex-dispatch.cmd) and one-line installers (install.sh,install.ps1). - Bilingual docs (English + Traditional Chinese README and CHANGELOG).
Install (one-line)
curl -fsSL https://raw.githubusercontent.com/fredchu/claude-codex-dispatch/main/install.sh | shPowerShell:
iwr -useb https://raw.githubusercontent.com/fredchu/claude-codex-dispatch/main/install.ps1 | iexSafety model
codex exec --dangerously-bypass-approvals-and-sandbox is always used. Safety comes from structure — task packet, write-scope check, git status snapshots, machine-readable verdict (policy.json), and orchestrator review — not from sandbox mode.
See CHANGELOG.md for the full list.