Consciousness Loop — Development skill for Claude Code

Consciousness Loop — a Claude Code skill

The **keystone** of the [agent-nervous-system](https://github.com/thdelmas/agent-nervous-system) suite. The other organs — [perception](https://github.com/thdelmas/open-source-octopus-investigation), [memory](https://github.com/thdelmas/rem-sleep), [defense](https://github.com/thdelmas/immune-check), [grief](https://github.com/thdelmas/sunset) — are *reflexes*: triggered, run, done. On their own they're a body with no pulse. This is the pulse: a **self-firing executive loop** that wakes on its own, pulls the organs into one workspace, decides what needs attention now, and schedules its own next wake.

It's the difference between an organism and a well-organized corpse.

Two things, paired

• The loop — a global-workspace cycle (wake → sense → integrate → decide → act/defer → sleep). The agent's self-initiation: it acts with no human prompt in the loop.
• The frequency — adaptive arousal. Alert (cache-warm, ~60–270s) when something's happening; drowsy (~20–30 min) when idle; deep sleep (long / event-only) when nothing's expected. The governing law: never tick faster than the world you're watching changes.

It's the **wake phase** to rem-sleep's **sleep phase** — the loop is what decides when to sleep. Two arcs of one circadian cycle.

Functional, not phenomenal

This is a global-workspace integration loop and an arousal clock — **not** a claim about subjective experience. The hard problem is out of scope; continuity, self-initiation, integration, adaptive cadence, and knowing when to stop are in scope. Read "consciousness" as *functional executive*.

Requirements

A **wake scheduler** — a self-scheduling primitive (a `ScheduleWakeup`-style timed re-invocation), a recurring runner (`/loop`, cron, a job queue), or, best, an **event hook** that re-invokes the agent only when state changes. With none, the loop degrades to a single pass plus a recommended next-wake the human triggers.

Install