Yomi

Appearance

Agent Skills ​

Yomi can install command guidance as project-local agent skills through the Crust skills plugin.

bash
npx @isamisushi/yomi@latest skill --all --scope project

In this repository:

bash
npm run yomi -- skill --all --scope project

For an installed CLI:

bash
yomi skill --all --scope project

Included Skills ​

Yomi currently ships:

  • yomi - generated command guidance for the CLI.
  • yomi-react-repair - how an agent should use repair, evidence trails, and do-not-start hints during React bug work.
  • yomi-react-instrumentation - how an agent should use plan-trace and instrument to collect runtime history without tracing the whole app.

Generated vs Bundled Skills ​

The yomi skill is generated from the Crust command tree. It is the best place for an agent to check exact command names, flags, defaults, and help text.

The React skills are bundled workflow skills:

  • yomi-react-repair tells the agent how to use repair briefs without blindly editing display-only components.
  • yomi-react-instrumentation tells the agent when to add focused runtime trace points and when to avoid broad instrumentation.

Use both kinds together. The generated skill answers "what command exists?" The workflow skills answer "when should the agent use it?"

Install Location ​

With --scope project, Yomi writes skills into project-local agent directories. Current Crust skill installation writes the universal agent skill path and any detected agent-specific paths, such as:

txt
.agents/skills/
.claude/skills/

This keeps the guidance inside the repository instead of relying on every developer's global agent setup.

Use project scope for OSS examples, team repositories, and benchmark fixtures. Use global scope only for your own machine:

bash
yomi skill --all --scope global

Update Skills ​

Re-run installation when the CLI changes:

bash
yomi skill update --all --scope project

Do this after upgrading @isamisushi/yomi, changing command names, or changing the bundled workflow skills. Otherwise an agent may follow stale flags or stale workflow rules.

Agent Setup Flow ​

For a new repository, the recommended setup is:

bash
npx @isamisushi/yomi@latest index --force
npx @isamisushi/yomi@latest doctor
npx @isamisushi/yomi@latest skill --all --scope project

Then commit the generated project-local skills only if you want every coding agent working in the repo to share the same Yomi workflow guidance.

Why Skills Matter ​

The CLI output is JSON, but agents also need operating discipline. The skills encode rules such as:

  • start from the visible symptom
  • trust editTarget as the first inspection point, not a blind patch location
  • treat doNotStartFrom as evidence-only unless source inspection proves the graph is stale
  • use plan-trace before choosing broad instrumentation
  • verify the visible behavior after editing

This keeps Yomi from being just another command list. It gives an agent a repeatable repair workflow.