Skip to content
doraval

Quickstart

Scale context for yourself, your team, or your community. Same workflow whether you are tuning your own agent, onboarding juniors, or publishing a plugin.

This quickstart assumes doraval is installed. If not, see Installation.

  Scale AI context (you, team, or community)
                    │
                    ▼
             ┌─────────────┐
             │  claude new │  scaffold skill or plugin
             └─────────────┘
                    │
                    ▼
             ┌─────────────┐
             │dora validate│  catch breakage before anyone uses it
             └─────────────┘
                    │
                    ▼
             ┌─────────────┐
             │  dora init  │  journal + coding agent
             └─────────────┘
          ┌────────┴────────┐
          ▼                 ▼
   ┌─────────────┐   ┌─────────────┐
   │journal flow │   │  dora eval  │
   │ add · sync  │   │ real sessions│
   │ hook enable │   │ ADHERES/DRIFT│
   └─────────────┘   └─────────────┘
          │                 │
          └────────┬────────┘
                   ▼
        context that works first try

  1. Scaffold context

    Section titled “Scaffold context”

    Create a skill or plugin for the agent you use day to day:

    Terminal window
    # Personal / team context
    doraval claude new --yes --intent self my-agent-context
    

    # Plugin to ship to others
    doraval claude new --yes --intent distribute my-plugin

    Use doraval cursor new, doraval codex new, or doraval copilot new for other agents. See claude new.

  2. Validate before you rely on it

    Section titled “Validate before you rely on it”

    dora validate auto-detects what you scaffolded and runs every check that applies: skills, plugins, hooks, MCP servers, subagents, memory files, and more.

    Terminal window
    dora validate .
    dora validate . --for claude:plugin
    doraval providers

    See validate and providers for the full matrix.

  3. Set up doraval and your coding agent

    Section titled “Set up doraval and your coding agent”

    Run dora init once to wire up your journal and the agent doraval uses for enrichment and evals:

    Terminal window
    dora init

    Or set the agent directly:

    Terminal window
    dora config set agent.command claude
    dora config get agent

    Claude Code and Grok have the deepest support today. Cursor, Copilot, and OpenCode work for scaffolding and validation, with fuller eval integrations expanding over time.

  4. Record decisions that stick

    Section titled “Record decisions that stick”

    Capture principles as you work so future sessions (and other people) do not rediscover the same choices:

    Terminal window
    dora journal add "Use validate before shipping any skill change"
    dora journal sync

    Install a SessionStart hook so your agent receives the journal before the first message:

    Terminal window
    dora journal hook enable

    See journal add, journal sync, and Agent Journal.

  5. Measure adherence in real sessions

    Section titled “Measure adherence in real sessions”

    validate checks the document. dora eval checks reality: did the agent actually follow your skill in a real transcript?

    Terminal window
    dora eval

    Pick a recent session interactively, or target one skill with judge. A DRIFTED verdict means the skill did not hold in practice — catch that before you scale context to more people.

  6. Scale evals across sessions Coming soon

    Section titled “Scale evals across sessions ”

    When context reaches a team or community, you care about patterns across many sessions, not just one run.

    Terminal window
    dora eval --all --parallel

    Until then, run dora eval per session and use dora eval history to track drift over time.

Next: explore command reference pages for drift, ui, and update.