What is doraval?

Scale your AI context for coding agents. Make your next context work (skills, plugins & more) for your team, community, or self. Context engineering toolkit for AI coding agents.

Whether you are optimizing context for yourself, your team, or a wider community, the job is the same: make skills, plugins, and decisions work on the first attempt — not after days of debugging.

Quickstart →

Name and pronunciation

doraval is pronounced dor-uh-val — stress the last syllable, like Dora + eval.

The Dora comes from Doraemon, the robot cat who pulls the right gadget for the job. The -val comes from evaluation — session evals that judge whether an agent actually followed your skills, not just whether the file parses. validate and skill drift check the document; eval checks reality. The shorter alias dora is the same CLI.

The problem

Give 10 engineers (or agents) a skill. Only 3/10 succeed on the first try.
Publish a plugin to the community? Most adopters hit the same slow ramp.
Decisions drift across sessions because nothing persists.
You cannot confidently say “this works for Claude, Cursor, and Copilot.”

doraval left-shifts success: catch breakage before the first session, not after the third debugging thread.

How it fits together

  Scale AI context (skills, plugins, decisions)
                        │
                        ▼
                 ┌─────────────┐
                 │   doraval   │  validate · scaffold · journal · eval
                 └─────────────┘
                        │
                        ▼
        yourself · your team · your community
                        │
                        ▼
              first attempt succeeds

What doraval does

Command	What it scales	Speed
`validate`	Catch broken skills, plugins, hooks, MCP before anyone uses them	Instant
`claude new` / `cursor new` etc.	Scaffold resources that load correctly across agents	Instant
`init`	One-time setup for journal + coding agent	Instant
`journal`	Decisions that persist across sessions and people	Instant
`eval`	Proof that agents followed your skills in real sessions	Minutes
`skill drift`	Rubric deviation before subjective failure shows up	Instant
`ui`	Local dashboard for capturing context without retyping	Instant
`providers`	See which agents understand which packaging	Instant

validate is the workhorse. Point it at a directory or GitHub URL and it tells you what is broken before you scale context to anyone else.

What to do next

Install doraval
Quickstart — scaffold, validate, journal, eval in one path
doraval update — stay current
doraval providers + --for — target the right agents

The goal is always the same: scale context that works on the first attempt.