cd /news/ai-tools/koordinate-thinking · home topics ai-tools article
[ARTICLE · art-137358] src=github.com ↗ pub= topic=ai-tools verified=true sentiment=· neutral

Koòrdinate Thinking

UnmarkedPM released Koòrdinate Thinking as a self-contained FlatSkill file in Italian and English, each 24 KB, requiring no platform-specific packaging. The Markdown file carries the full method in context with no lazy loading, unlike the OpenAI and OpenWebUI packagings that depend on a skill directory or native function calling. The release states the procedure prevails over the condensed rationale in section 7, and delivery checks rest on the model's judgement with no executable verification.

read3 min views1 publishedSep 22, 2026
Koòrdinate Thinking
Image: Michielbdejong (auto-discovered)

The method as one self-contained file, with no packaging around it. One file per language:

[`koordinated-thinking-flatskill-it.md`](https://github.com/UnmarkedPM/koord/blob/main/tools/skills/FlatSkill/koordinated-thinking-flatskill-it.md) (24 KB) and
[`koordinated-thinking-flatskill-en.md`](https://github.com/UnmarkedPM/koord/blob/main/tools/skills/FlatSkill/koordinated-thinking-flatskill-en.md) (24 KB). Install the one

whose language you work in — the two are the same instrument, and which language is which is not

incidental; see Italian and English. The other two packagings assume a platform: OpenAI/ needs a client that reads a skill directory, OpenWebUI/ needs a skill store and native function calling. A FlatSkill assumes nothing. It is Markdown with no front matter, no companion files and no references to a file system, so it goes wherever an assistant will accept a block of instructions:

  • a system prompt, or the instructions field of a project, a custom GPT, a Gem, an assistant;
  • a file attached to a conversation, on a platform that reads attachments;
  • the first message of a chat, on a platform that offers nothing else;
  • an ordinary document, for a human being who wants to practise the method without a model at all.

It is also the form to start from when adapting the method to a platform not covered here: one file to read, with nothing resolved elsewhere.

Section Content
1 The procedure: the two phases, when the method applies, opening, the four-phase working cycle, the drafting thresholds, delivery
2 The actors: discipline manager, Critic with the full checklist and intervention formulas, thematic experts, and the declared limit
3 The discipline — the good-enough default and how to raise the threshold
4 Additivity and the prohibition on loss, with the prohibited moves named one by one
5 The document model: chapters 0–3 and annexes A–F, with the fixed text of the sections that carry it
6 Working principles
7 The rationale in brief, with declared limits and what the method does not claim

Sections 1–4 are operational and are to be followed, 5 is the model to fill in, 6 and 7 serve for cases the procedure does not foresee and for answering questions about the method.

Everything is in context, always. There is no lazy here: the whole ~24 KB sits in the prompt for the entire conversation, where the Claude skill pulls a reference file when it needs one and the Open WebUI package pulls a companion skill with view_skill. On a small context, or where the prompt is billed at every turn, that is the cost to weigh.

The rationale is condensed, not carried whole. Section 7 states the four pillars, the declared limits and what the method does not claim, in about a page; the other packagings ship the method document itself as their rationale layer, some 48 KB. Where section 7 and the published document appear to diverge, the file states the precedence: the procedure prevails, and the divergence is flagged to the operator.

There is no executable verification. The delivery checks rest on the model's judgement — true of the Open WebUI package too, and unavoidable in any packaging that is only prose.

Paste or attach the whole file. Nothing has to be edited first, and nothing outside it has to be reachable. If the environment has a file system or a writing tool the procedure will use it, saving each version as <project-name>-vX.Y.md; if it has neither, the document is delivered as a Markdown block in the conversation and keeping it is then the operator's job — one note or one file per project, replaced whole at each version, pasted back when resuming the work. The procedure asks for the specification rather than reconstructing it from an earlier conversation, because a specification rebuilt from memory is not a source of truth.

CC BY 4.0, SPDX CC-BY-4.0, like every skill in this repository whatever it packages for. The notice is at the foot of each file, which matters more here than anywhere else: a file meant to be pasted travels further from this directory than any other, and has to say what it is on its own. Attribution text: ../../../method/ATTRIBUTION.md. The authoritative map for the repository is ../../../LICENSING.md.

── more in #ai-tools 4 stories · sorted by recency
── more on @unmarkedpm 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/koordinate-thinking] indexed:0 read:3min 2026-09-22 ·