Codex adaptation and security boundary

This is a per-turn native Codex guidance adapter. It preserves the 23 playbooks and their principles, but it does not create a sticky mode or claim implicit future invocation. Read the selected playbook from the project-local mirror. Map the upstream question action to the host's optional user-input surface only for a genuine product or preference choice. Map the upstream task and delegation fields to native bounded delegation when exposed, keeping scope, selected model/effort, ownership, and proof with the coordinator. If named delegation is unavailable, report the exact gap; sequential local work may continue, but it does not complete the named-role playbook step. For cleanup, use cursor-deslop when installed; for CLI/TUI surfaces, use cursor-control-cli; for browser/Electron/web surfaces, use cursor-control-ui. Check each concrete skill before use and record the exact gap when unavailable or not applicable. Cloud-capable task environments are conditional; if the cloud lane is unavailable, record the gap and do not claim parity. Keep model selection with model-capability-router and proof-path selection with proof-orchestrator. A heartbeat or automation is used only when explicitly requested. Ordinary playbook delegates call native collaboration.spawn_agent with agent_type: "poteto-agent" when that exact type is exposed; reuse an existing suitable named agent with collaboration.followup_task instead of spawning a sibling. The profile is distributed by model-capability-router; install it explicitly before a fresh Codex session can select it. If the exact type is unavailable, report the gap and leave named-role execution incomplete rather than substituting a generic agent. Read the selected cursor-principle-* leaf when applied. Routed workflow skills (cursor-how, cursor-why, cursor-interrogate, cursor-reflect, cursor-swarm, cursor-architect, and cursor-arena) keep their own role-specific contracts. Reuse the same named delegate when supported. Automatic sticky routing, fresh top-level profile loading, and full live playbook execution remain unproven; native selection and bounded follow-up of a named poteto-agent have been observed in this task tree.

This skill is published and indexed for explicit invocation. Installing it makes the guidance, playbooks, references, and reviewed profile dependency available; it does not authorize the bundled Bun/TypeScript CLI, dependency installation, real-repository mutation, or GitHub access. Those executable surfaces remain separately gated. The detailed fixture and security evidence stays in MIRROR.md; it is historical proof, not an active installation prohibition.

Poteto owns principle selection inside its playbooks through the inline index below. Read each selected leaf in full. Do not run a second principle-selection pass through principle-pack-router inside this workflow; the global router remains available for choosing workflows elsewhere and operator policy remains authoritative.

Poteto mode

Bundled skills live at ../<name>/SKILL.md from this file, so put this file's absolute path in every poteto-agent prompt.

Non-negotiables

The Principles section below grounds every trigger. In your reply, name each principle that shaped a decision and the specific choice it changed. Cite only principles whose leaf SKILL.md you read this session.

Remaining triggers: