Shann³(@shannholmberg):a company skill library needs 5 layers keep every approved skill in one shared library, agents check it at session start or on a schedule, then update their local copies automatically 1. source of truth > one shared GitHub repository > every approved change is committed to its history > every skill has an owner, version and last update 2. discovery > a searchable internal page shows what exists > each skill explains the task it runs, when to use it and the output it produces > people search by outcome, team or workflow 3. loading > the agent reads the skill when the task matches > it also reads the relevant company files, brand rules and examples > the skill defines the steps, checks and output format. the company files provide the facts 4. improvement > someone runs the skill on a campaign, client brief or report > they fix the missed step, weak output or instruction about which source to trust > the correction goes back into the shared version > permitted people and agents receive it when their files update 5. governance > changes are reviewed before they reach everyone > old versions stay recoverable > overlapping skills get merged, unused ones get retired > a permissions file defines which teams and agents can read restricted skills

2026.08.26 19:14

a company skill library needs 5 layers keep every approved skill in one shared library, agents check it at session start or on a schedule, then update their local copies automatically 1. source of truth > one shared GitHub repository > every approved change is committed to its history > every skill has an owner, version and last update 2. discovery > a searchable internal page shows what exists > each skill explains the task it runs, when to use it and the output it produces > people search by outcome, team or workflow 3. loading > the agent reads the skill when the task matches > it also reads the relevant company files, brand rules and examples > the skill defines the steps, checks and output format. the company files provide the facts 4. improvement > someone runs the skill on a campaign, client brief or report > they fix the missed step, weak output or instruction about which source to trust > the correction goes back into the shared version > permitted people and agents receive it when their files update 5. governance > changes are reviewed before they reach everyone > old versions stay recoverable > overlapping skills get merged, unused ones get retired > a permissions file defines which teams and agents can read restricted skills

Shann³@shannholmberg

2026.08.25 06:31

better idea (how to create and maintain a company wide skills library) train your team on git or something agent-friendly downloading skills through a UI every time there is an update kills workflow, instead, set up auto-scan or hook agents to always pull the latest version of the skills library then train the team to use skills made by others and build a culture where everyone share what they learn here is how the setup works: > store the skills library inside your company brain (Github repo for example) > train the team on how to use the brain, not just search for skills but also load prio context you have saved in there > when someone improves a skill, the update goes to the library and everyone gets it on the next agent interaction > skills get better every time someone uses them and logs what worked to view the library, create a simple HTML page that lists every skill with its use case and last update