coding-agent-toolkit
A cross-agent workflow toolkit of seven steps for taking work from idea to merged PR, usable as Claude Code skills or Codex prompts, with shared on-disk state so either agent can pick up a task.
coding-agent-toolkit is a set of seven workflow steps, invoked as slash commands in Claude Code or as prompts in Codex, that structure the work around building a feature rather than the coding itself: defining what done means, breaking work into chunks, projecting a plan onto GitHub for review, and carrying context across sessions and hosts.
The align step surveys the codebase read-only and interrogates the user into a spec file; plan breaks the spec into dependency-ordered chunks; plan-issues turns those chunks into a GitHub milestone and issues, dry-run by default; next-issue takes the lowest-numbered open issue in the milestone through implementation, tests and a pull request; and ship closes the milestone. handoff and continue write and replay a snapshot of read-only checks so work can resume after a context limit, restart, or day boundary. Every step ships as both a Claude Code skill and a Codex prompt reading and writing shared state under a workspace directory, so either agent can continue where the other left off.
It is written in Shell and MIT licensed.