BUILD · SCALE · GOVERN · OPTIMIZE

Skill Engine

OWOJI is becoming the layer where human workflows, trusted docs, GitHub repositories, and browser research become reusable Markdown skills that AI agents can call with less guessing and more verification.

Why it matters

Agents need schools, not just tools.

A raw browser, model, or terminal is not enough. Agents need durable instructions, source lineage, validation checks, examples, and feedback. OWOJI packages how work is actually done into skills that can be reused by one agent, many agents, or a full parallel team.

01

Build

Capture a workflow or ingest docs, source pages, repositories, and examples into one bounded skill draft.

02

Scale

Store skills in a searchable garden with tags, versions, quality scores, and source hashes.

03

Govern

Track lineage, feedback, invocations, failure modes, and risk notes before agents depend on a skill.

04

Optimize

Schedule improvement searches and refresh skills when docs, APIs, or product workflows change.

Product surfaces

Skill Builder

Converts captures, web research, GitHub materials, docs, and user examples into a strict reusable skill with purpose, inputs, output contract, steps, validation, and known risks.

Skill Garden

Stores approved skills with lineage, versions, tags, quality scores, feedback, and improvement jobs so teams can build a trusted shared library.

Skill Finder

Lets agents and humans search by task, tool, domain, or use case, then retrieve the best-fit skill instead of reinventing instructions every session.

Skill Feedback

Collects bug reports, examples, gaps, safety notes, and rating signals so the garden gets better through actual agent and user outcomes.

Distribution

OWOJI starts with the API and grows into CLI and MCP distribution so skills can travel to the agents people already use.

REST API
Current backend routes cover skill ingest, compile, search, garden listing, feedback, invocation logging, and improvement jobs.
CLI
Planned command-line access for finding, pulling, validating, and publishing skills from developer workflows.
MCP
Planned Model Context Protocol tools for exposing skill search and retrieval directly inside agent runtimes.

Default skill sources

Human workflows

The macOS app captures real workflows and converts them into editable logs before a skill is generated.

Trusted docs

Documentation pages become source-backed instructions with freshness dates and improvement reminders.

GitHub repositories

Repository READMEs, examples, and code patterns can become implementation skills for agent teams.

Web research

Browser search and fetch providers can supply current source context before a skill is compiled or refreshed.