Reusable, governed procedures for orchestrating AI agents — the methods & procedures pillar of Agentic Process Engineering.
Each procedure ships as a Claude Code skill. The principle is the same one
APE applies everywhere: the process is the durable asset, the tool is
swappable. Today these procedures run over acpx
with Codex as the worker; the transport underneath can change without the
procedure changing. You keep the process; you keep it yours.
The shared shape across every procedure: Claude orchestrates and judges, an external worker (Codex by default) executes. Claude frames the briefs, dispatches, reconciles, and synthesizes; it does not type the labor.
| Skill | Shape | Use for |
|---|---|---|
pair-codex |
one worker, sequential | TDD pair (red→green→refactor), or a simple fan-out |
acpx-workflows |
N workers, parallel | multi-axis review/audit, find→verify, adversarial verify, loop-until-dry |
Both are the external-worker mirror of Claude Code's native orchestration:
pair-codex ≈ the Agent tool (one sub-agent, sequential), acpx-workflows
≈ the Workflow tool (parallel fan-out). Same patterns — the worker is
Codex instead of Claude, so the labor lands on an agent you already pay for
elsewhere instead of burning Claude tokens.
These procedures use the acpx transport
skill (skills/acpx in that repo). Install it first, then:
git clone https://github.com/Casys-AI/agentic-procedures /tmp/agentic-procedures
cp -r /tmp/agentic-procedures/skills/pair-codex ~/.claude/skills/
cp -r /tmp/agentic-procedures/skills/acpx-workflows ~/.claude/skills/In any Claude Code session: "pair with codex on this bug" triggers
pair-codex; "fan out a codex review across architecture, perf, security"
triggers acpx-workflows.
A procedure that names a tool dies with the tool. APE separates the layers
on purpose: the method (how Claude frames, dispatches, judges) is engineered
once and outlives the tools & systems beneath it. When a better transport
than acpx appears, the briefs, the anchors, and the quality patterns carry
over unchanged.
© 2026 Casys AI. Licensed under CC BY-SA 4.0 — same license as the APE canvas. Reuse and adapt freely, with attribution; derivatives stay under CC BY-SA.