In a session today there was some discussion about agentic skills. A skill is mostly a markdown text file often named, yes, skill.md with a natural language text plus some front matter, describing some specific workflow or bit of knowledge or capability.
Somebody external opined that such skills were a good way of codifying organisational knowledge, which might be useful ‘because of the ageing demographics’ on the work floor and people retiring.

Would agentic skills succeed where process descriptions, intranets, 1990-2010 knowledge systems, quality assurance systems all failed when aimed at documenting the actual knowledge present in an organisation?

I doubt it.

Deep knowledge is relational, chunky, only its underlying facts and heuristics can be codified. And people will resist codification, for different reasons. Such as recognising the futility of it, or recognising its aim of making the people with knowledge superfluous. Or both.

Are skills different enough from earlier systems aimed at codification to gain some willing cooperation from knowledge workers?

They might be.

Unlike the earlier systems, a skill is small and its aim is not explicitly ‘codifying everything’. Smaller than you, an information object in itself. And you get to combine such objects into a chain or bundle as a small workflow. They are a much smaller ask than the others were.
Additionally skills can be deployed by yourself in your own work. For processes, QA, KM it was mostly all done for others, for the hand-offs between actors and mostly seen as external to one’s own work. Why would you write down and maintain what you would not use yourself?
That is where skills can have a different impact. And in terms of ease of use, it helps that such skills are in natural language, as it allows you to handwave the parts where you can’t express your own actual knowledge in any precise detail. Which will be often.

So skills will be likely an easier sell to do some codification.
Still the push for skills is to be small, niche and specific. I’ve seen lots that are actually deterministic, in other words can be automated, except perhaps for the decision when to call a skill and with which inputs. Meaning they codify something different than knowledge.
Such codification will not likely form a coherent whole either. Also because skills are regularly aimed at something that is not within the scope of existing knowledge, a helper function next to current knowledge work.

So, no, I don’t buy the ‘skill codification as knowledge transfer’ suggestion to sell skills with.