Executable markdown
A markdown document whose code blocks run — the note stops being a record of a computation and becomes the computation. Named here from yank-note, the first tool in this wiki's roster to do it.
Where it sits in the lineage
Every prior step in this corpus made notes more connected: zettelkasten made them atomic and linked, the memex made the links traversable, obsidian and roam-research made them a graph, the llm-wiki pattern made a machine maintain them. All of it operates on documents that are inert — read, rendered, linked, never executed.
Executable markdown adds a different axis. The document acquires effects: it can read files, call the network, run a shell. Jupyter and the literate-programming tradition got there first for code-centric work — and reproducible-research got there for a second reason, so a stranger could re-run a published analysis. What is new is the property arriving in a personal knowledge base, where the documents come from everywhere and are opened casually.
The trust boundary it creates, and the corpus's exposure to it
yank-note states the trade instead of hiding it: extensibility is bought by giving up protection against command execution and arbitrary file read/write, so an untrusted markdown file is dangerous to open.
That matters here beyond one product, because of what the rest of this wiki describes. The llm-wiki pattern has an LLM writing markdown continuously, often from web sources it fetched; gbrain runs the same loop unattended. In that pipeline, prose and code arrive by the same route and carry the same trust level — which is none. An executable document turns a summarization pipeline into an execution path, and the augment→automate axis makes it worse toward the automated end, where no human reads the file before it is opened.
The mitigations are the ordinary ones: run untrusted documents in a sandbox, require an explicit gesture per execution, or keep the reading tool and the executing tool separate — which is what a plain-markdown setup does by accident.
Corrected 2026-08-14. This page originally said none of the three was evidenced anywhere in the
corpus. The middle one is. org-mode-literate-programming documents a per-block confirmation
prompt as org-mode's default since 2012 (org-confirm-babel-evaluate), and argues for it on a
ground this page independently arrives at: the more languages a document mixes, the less able the
reader is to audit what they are about to run. So the gesture is not an untried idea — it is the
oldest executable-document system's answer, thirteen years old. The sandbox and the tool separation
are still unevidenced here, and nobody has measured whether a confirmation prompt survives contact
with a user who sees it fifty times a day.
Open
- Nobody in this corpus has written down a threat model for LLM-authored notes. The tools assume the author is you.
- Whether the productivity gain is worth the surface is unmeasured, in both directions: no incident reports, and no evidence that executable notes make anyone more effective.
Related
yank-note · tools-for-thought · llm-wiki · obsidian · zettelkasten · gbrain · literate-programming · reproducible-research · org-mode-literate-programming