Definition
A skill library is the collection of skills a team maintains as reviewable, versioned artifacts — each with a type, a visibility scope, and a record of the sessions it came from — so that skills can be found, governed, and reused rather than scattered across individual machines and chats.
A skill library is the collection of skills a team maintains as reviewable, versioned artifacts. One extracted skill helps one run. A library is what turns a growing set of extractions into something a team can search, review, and rely on — the difference between a useful pattern living on one person’s machine and the same pattern being available to everyone.
Quick breakdown
A library is more than a folder of files. Each entry carries structure that makes it findable and governable.
| Type | Whether the skill is a workflow, a prompt template, or domain knowledge, so procedures are distinguishable from reference material. |
|---|---|
| Version | A version on each skill, so changes are deliberate and a prior version can be restored. |
| Visibility | A scope of private or team, so a skill can mature before it is shared. |
| Provenance | A record of the sessions a skill came from and whether it was AI-generated. |
| Tags | Labels that make a skill findable by task, surface, or domain. |
Why a team needs a skill library
Without a library, skills are private events. One engineer extracts a useful pattern, and it stays on their machine or in a chat thread. The next person solves the same problem from scratch.
A skill nobody else can find is a solved problem the team will solve again.
A library changes the unit of value from the individual run to the team. A pattern one person discovered becomes a pattern anyone can pull up, see the origin of, and apply. The library is the structure that lets individual problem-solving accumulate into shared capability instead of evaporating.
How skills are organized in a library
Libraries stay usable when entries are typed and labeled rather than dumped together. A skill carries a type — a workflow, a prompt template, or domain knowledge — which keeps step-by-step procedures distinct from reference material an agent reads for context. Tags add a second axis: by task, by tool surface, by domain. A slug and a name make each skill addressable.
The result is a library you can navigate by intent. “Show me the workflows for our deploy tooling” is a question a typed, tagged library can answer. A flat pile of text cannot.
How a library stays trustworthy: review and versioning
A library is only as good as the trust a team places in it, and that trust comes from two habits.
New or edited skill
│
▼
Review (a person reads it, checks the steps)
│
▼
Version bump (the change is deliberate, and reversible)
│
▼
Visibility (private while it matures → team once trusted)
│
▼
Available to future runs Review before adoption. Because a skill shapes later runs, adding or changing one is treated like a change to any shared artifact: someone reads it and decides it belongs. Versioning makes each change deliberate and reversible, so a skill that drifts can be rolled back rather than silently trusted. Visibility scoping supports both, letting a skill stay private until it has earned a place in the shared set.
How a skill library differs from a prompt library or RAG store
A skill library is sometimes confused with two adjacent things. The difference is in what each one stores and how it is maintained.
| Collection | What it stores | How it is maintained |
|---|---|---|
| Prompt library | Hand-authored instructions | Edited by hand; little provenance or structure |
| RAG store | Reference documents for retrieval | Indexed and re-indexed; reference, not procedure |
| Skill library | Typed, versioned procedures with provenance | Reviewed and versioned like a shared artifact |
The collections are complementary. A team can keep all three. What the skill library adds is the procedural layer with a record of where each entry came from — the part a prompt list and a document store both tend to leave out.
What a growing skill library makes possible
The value of a library compounds. The first skill is a curiosity. By the time a library holds the team’s recurring procedures, a few things change at once.
- Onboarding shortens. A new engineer inherits the team’s extracted procedures instead of rediscovering them.
- Behavior converges. Runs route through the same reviewed skills rather than each person’s separate guess.
- Knowledge stops walking out the door. When a pattern lives in the library, it survives the person who first found it.
How Paper Compute stores and shares skills
In paper console, each skill is a stored artifact with a type, a version, a visibility scope of private or team, and a record of the sessions it was generated from. That makes the library searchable and governable in one place. Teams that prefer their skills in version control can also keep them as files in a repository, where changes to a skill are reviewed like any other change to code. Either way, the library is the shared layer that skill invocation draws from when a run needs a skill, and the foundation of team-shared agent knowledge.
A skill library is how a team stops re-solving its own solved problems.