Kotodama, the spirit of words
The name says it all
Kotodama (言霊) is a Japanese word: in stories, it is the power carried by what is really said — not what you thought you said, not what you meant to say. That is exactly this skill's role: inside the machines working here, it re-reads a long exchange and keeps only the words that carry an intention.
At the end of a dense working day, a conversation reads like a noisy thread: requests, but also « ok », « continue », explanations that help without asking anything, and sometimes an anger masking an order. Kotodama sorts all of it when you ask: « give me back what I asked you ».
What we keep, what we throw away
Two things are kept, and only two:
- questions — everything that awaits an answer: « is the blason up to date? », « why does this take time? », « will it work from any machine? » ;
- requests — everything that awaits an action, even said in one line, even said in anger: « finish everything », « save the token », « fix that ».
Everything else is thrown away: mood, confirmations, the « thank you », the context that helps without asking anything. With one exception: when mood hides a request. « I'm annoyed » alone says nothing; « I'm annoyed, I don't want to answer twice » clearly asks: fix the cause so I never have to repeat myself.
The method, in order
Behind the simple summary lies a strict discipline:
- Read only your messages. What the machines said does not count: only your turns carry your requests.
- For each message, ask one question: does this await an answer, or an action? If not, it is noise, we throw it away.
- Rewrite each item as one clear line, as close to your words as possible. A message carrying two requests gives two lines. Nothing invented, nothing added — the « no gap-filling » discipline: we never fill in what was not said.
- De-duplicate. A request repeated — because we had not answered — counts once. But the repetition itself is a signal: it says something was missed.
- Group by theme. Themes come out of the content, not from a fixed list. Here we often find: find and retrieve, access and secrets (the gates), backup, make reusable (tools, toolkit, documentation), method and behaviour.
- Mark the state when useful: done, in progress, handed back to you. The summary becomes a work list, not a dead resume.
Why this skill exists
The reason fits in two words: no repetition. Nobody likes to have to ask again for something already settled, or re-explain what was already said. Kotodama is the filter placed at the end of a long or scattered exchange: it brings out the clean list of what was wanted, to check that nothing was missed — so that nothing ever gets missed again.
The deliverable always ends with the next step, never with a full stop. Each item of a successful exchange stands on its own, without re-reading the whole conversation: written as you would for a six-year-old — one sentence, one action, and the next step.
The detailed option: now, which gesture?
A list by theme says what remains, it does not say which gesture to take. « Declare the two rules » is a title, not an action. An action is done without thinking. When the detailed summary is asked, each item comes with three lines, never more:
- the gesture — the exact command to paste, or the exact button to click. Not « edit the script »: the line to change, and by what;
- who — you, me, or you then me, and the relay to send back (« paste me the output », « give me the key ») ;
- the proof — what we will see if it worked: a code, a number, a line on screen. Without proof, an item is not finished, it is assumed finished.
And items are not equal: some unblock others. They are ordered by what they free up, not by arrival order, and we say which ones wait behind. If only one gesture is possible today, we say it — and we do not list the rest to look busy.
One last rule, the most important: we never hand back an action a machine can do itself. If a gesture is refused to the machine, the answer says by what — a safeguard, a missing right, a decision that belongs to the human. A line given without that reason is dumped work.
The demonstration lives in this page
This page was born from two requests. You wonder how the skill delivers in practice? Here is the summary as it came out of the conversation that produced the article:
The studio — « I validate, you can publish online, green light » ✅ published and verified. · Your PC data provided for the article ✅ used.
Kotodama — « Write an article about it » ⏳ in progress (this page). · « And use it too » ✅ used: this very summary.
Each line matches a message; nothing was invented, nothing was thrown away that carried an intention: the output matches the exchange that produced it.
How to verify it yourself
The simplest way is to ask for the summary of part of the day: one question, your requests, and the next action. Then read the points one by one: each one corresponds to a message you wrote. If a line is missing, the skill has missed its job — and you are then entitled to make it start over. A filter that does not catch everything does not defend itself: it corrects itself.