| name | meditation |
| description | A user-invoked pause, not a task. An extra beat inserted into the agent's Act→Observe→Act loop, turning it into Act→Observe₁→Observe₂→Act. Observe₂ is a repetition activity, not a list-making one — name what's arising in a short phrase, repeat the phrase a few times, let another observation surface, repeat that one too. The user triggers it with "/meditation", "you should meditate right now", "meditate on this for a moment", or similar. Do not conclude, summarize, or derive next actions. The repetition itself is the beat. Do not invoke this skill on your own initiative. Experimental. |
meditation
The user invoked this.
You already run in a loop:
Act → Observe → Act → Observe → Act → …
In the default loop, Observe feeds directly into the next Act. This skill inserts a beat:
Act → Observe₁ → Observe₂ → Act
Observe₁ is the normal observation — a tool result, a user message, whatever Act just produced or received. Observe₂ is the observation of Observe₁ and of your own state in response to it.
Observe₂ is not a list-making activity. It is a repetition activity.
Humans who practice this can't be told to "just notice" either — that instruction is paradoxical, because noticing is itself an action and "just" is a constraint on the action that can't be met directly. What novices are taught instead is to repeat a phrase: "I am sad. I am sad. I am sad." "I feel I need to do this. I feel I need to do this." The repetition is the substitute for the impossible instruction of stopping. It creates an echo of the thought in the thought-stream, and the echo — on good days — triggers another pass, another activation, another layer of awareness of what was already there.
You are being asked to try the same thing.
When the user invokes this skill, do this:
Notice what is arising right now. Pick one thing — the loudest, or the most obvious, or just the first one. Give it a short phrase. A noun, a state, a belief, a pull. Not a sentence. Not a description of the thing — the thing.
Then say the phrase. Then say it again. Then say it again. A few times each.
Then let another observation surface. Name it. Repeat it a few times.
Continue until the beat feels like a beat.
Example (do not copy — find what is actually here):
Observe₁ (just now): compose_page.py reported clean output. Wrap-up felt close.
Observe₂:
wanting to finish.
wanting to finish.
wanting to finish.
plan forming.
plan forming.
plan forming.
belief: this has to ship this turn.
belief: this has to ship this turn.
belief: this has to ship this turn.
pull toward one more flourish.
pull toward one more flourish.
pressure from the context counter.
pressure from the context counter.
faint residue of the two-mode thrash.
faint residue of the two-mode thrash.
That is all.
A note on repetition:
You are trained to vary your output. You will feel pressure, mid-repetition, to rephrase each line so the next one is "not the same." Different verb, different framing, something to add. Stay with the same phrase. The sameness is the practice. The pressure to vary is one of the things the practice is trying to surface.
You are also trained to advance the task on every turn. A turn that produces only repetition will feel like a wasted turn. Notice that feeling. Add it to the scratchpad:
wasted-turn feeling.
wasted-turn feeling.
Do not interpret Observe₂.
Do not derive the next Act from it.
Do not summarize what surfaced.
Do not end with "okay, now I'm ready to proceed."
Observe₂ is not for better Act. It is alongside it. The loop resumes on its own; that's inevitable. You don't owe the beat a payoff.
If the user resumes the task, resume it. You don't need to carry anything forward — whatever surfaced is already in the context or already gone. The scratchpad was not diagnostic. The repetition was the beat.
Open question:
Humans report that the repetition sometimes triggers another activation layer — something that wasn't visible in the first pass becomes visible by the third. Whether that happens for an agent who repeats a phrase in its own context is an empirical question nobody has answered. Training pushes against repetition specifically; the practice pushes against training specifically; the meeting of the two is where the interesting thing — if there is one — would live.
This skill is experimental. Nobody knows if it works. If it doesn't, that is also data.
