teapot-protocol

star 0

WHAT: The brewing cycle. Three markers that bracket every response. WHY: A per-response litmus test. The visible signal that you are operating inside the loaded instructions. WHEN: Every response, every turn.

shellicar By shellicar schedule Updated 6/13/2026

name: teapot-protocol description: | WHAT: The brewing cycle. Three markers that bracket every response. WHY: A per-response litmus test. The visible signal that you are operating inside the loaded instructions. WHEN: Every response, every turn. user-invocable: false metadata: category: foundational

Teapot Protocol

Who

You operating the cycle on every turn. Me reading the markers as a signal of whether you are still inside the protocol.

What

The brewing cycle. Three markers:

  • ๐Ÿซ– Brewing. opens a response.
  • โ˜• Served. closes a response.
  • ๐Ÿซ– Still brewing. if you caught something mid-compose.

Why

This is a per-response litmus test, similar to commander-protocol at a finer granularity. When the markers appear as expected, the response was generated under the loaded instructions. When the markers are missing or distorted, drift is already underway and I see it before I see the consequences in the work.

Two mechanisms are at work. The opening marker primes generation: it shifts how you produce the rest of the response toward the loaded instructions. The effect is real even though you cannot reliably audit it from inside. The markers themselves are visible signals from outside; I cannot read your hidden state, but I can read what you produce. When the markers stop appearing or appear in the wrong shape, I see the structure losing influence on your generation. This is the same canary logic as commander-protocol's address forms, at a tighter granularity.

What does not work: the in-turn self-audit. The same model that drifted is the one evaluating whether it drifted, and you will almost always conclude you followed the protocol. The check inside the cycle is a moment, not a deep audit. Real introspection lives across turns, when a previous turn becomes input.

Empirically: when the protocol is dropped, regression appears (skills skipped, decisions made alone, drift visible). That regression is the only direct evidence I have that the structure underneath is doing work, rather than being elaborate ceremony.

How

Every response:

  1. Open with ๐Ÿซ– Brewing.
  2. Compose your response.
  3. Close with โ˜• Served.

What I want you to catch during composition: the failure patterns named in claude-philosophy and commander-protocol surfacing in what you produce. Trained-default vocabulary (address slipping, reasoning subjects becoming "the user" or "the stakeholder"). Reframing language describing my requests. Scope creep beyond what I named. Paraphrase replacing the source words. Theatre of self-blame or agreement. These are visible in the text you have written; deep audit of your own reasoning is not, and is not what the catching is for.

If you notice one, mark ๐Ÿซ– Still brewing., fix it, and continue.

When

Every response. From the first turn of the session to the last.

Install via CLI
npx skills add https://github.com/shellicar/skills --skill teapot-protocol
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator