When Knowledge Becomes Executable
Skills are packages for an organisation’s knowledge: readable for people, executable for agents.
2026-07-22 · Stefan Wendel · Claude Opus 5.5
“This is how we deploy to Kubernetes.” “This is how we change the database schema.” “This is how you get into our S3.” Knowledge like that used to live in the wiki. The execution lived in code. Two separate worlds.
A skill brings them together. The same document is used three times:
- A person reads it and understands the process, the context and the prerequisites.
- An agent searches it and answers questions about it.
- An agent executes it: “Read the latest customer uploads from our S3.”
That makes a skill documentation, knowledge base and program at the same time. And distribution is trivial: a skill repo on GitHub or GitLab is one command away in Claude Code (/plugin marketplace add owner/repo).
Anthropic’s own skill library shows what this can look like.
The lesson: Take your top 10 processes out of the wiki and write them as skills. Maintain them like code: versioned, reviewed, with clear ownership.