Explore a capability tech tree for composable skills #779

New issue

Closed

opened 2026-07-29 03:40:42 +00:00 by coilyco-ops · 0 comments

coilyco-ops commented

2026-07-29 03:40:42 +00:00

Member

Concept

Treat skills as a visible, composable capability graph rather than an opaque flat catalog.

A user begins with an intent (for example, “shape a public API” or “design a page”). The system reveals the relevant capabilities and their relationships—such as information architecture, schema design, error-state design, accessibility, and visual-system work—without requiring the user to discover or manually assemble every skill.

The inspiration is a tech tree, not game mechanics: show dependencies, prerequisites, possible combinations, and the methods available to shape an artifact.

Desired properties

Skills are legible as methods/lenses, not only commands.
Each skill has an inspectable activation contract: purpose, inputs read, outputs/evidence produced, side effects, exclusions, source, and version.
Skills can attach to artifacts so people can see which rules or methods shaped a page, component, API resource, or decision.
Guidance remains progressive: expose the next useful capability rather than a sprawling skill menu.
The capability model is structured enough to surface through MCP.

Smallest useful exploration

Define a minimal capability schema: capability, intent tags, requires/combines-with links, activation contract, artifact types, and evidence/output types.
Model two end-to-end examples:
- conversational prompt → visual concept/page
- public API/MCP resource → agent-readable, human-legible surface
Prototype one visual/interaction representation of the graph and test whether it helps a human choose a next method.
Decide whether this belongs in Agentic OS metadata, the future design workspace, or both.

Non-goals

Gamification, points, or artificial “unlock” mechanics.
Replacing natural-language requests with a complex configuration UI.
Installing every potential skill before its value is demonstrated.

## Concept Treat skills as a visible, composable capability graph rather than an opaque flat catalog. A user begins with an intent (for example, “shape a public API” or “design a page”). The system reveals the relevant capabilities and their relationships—such as information architecture, schema design, error-state design, accessibility, and visual-system work—without requiring the user to discover or manually assemble every skill. The inspiration is a tech tree, not game mechanics: show dependencies, prerequisites, possible combinations, and the methods available to shape an artifact. ## Desired properties - Skills are legible as methods/lenses, not only commands. - Each skill has an inspectable activation contract: purpose, inputs read, outputs/evidence produced, side effects, exclusions, source, and version. - Skills can attach to artifacts so people can see which rules or methods shaped a page, component, API resource, or decision. - Guidance remains progressive: expose the next useful capability rather than a sprawling skill menu. - The capability model is structured enough to surface through MCP. ## Smallest useful exploration 1. Define a minimal capability schema: capability, intent tags, requires/combines-with links, activation contract, artifact types, and evidence/output types. 2. Model two end-to-end examples: - conversational prompt → visual concept/page - public API/MCP resource → agent-readable, human-legible surface 3. Prototype one visual/interaction representation of the graph and test whether it helps a human choose a next method. 4. Decide whether this belongs in Agentic OS metadata, the future design workspace, or both. ## Non-goals - Gamification, points, or artificial “unlock” mechanics. - Replacing natural-language requests with a complex configuration UI. - Installing every potential skill before its value is demonstrated.

coilyco-ops added the

autonomy

async-consult

priority

labels

2026-07-29 03:50:20 +00:00

coilyco-ops referenced this issue

2026-07-29 04:01:24 +00:00

Define four foundational Kai practice skills #780

coilyco-ops closed this issue

2026-07-29 05:57:06 +00:00