# doraval

> Scale your AI context for coding agents. Make skills, plugins, and decisions work on the first try — for yourself, your team, or your community.

# Scale your AI context for coding agents
Source: https://doraval.dev/

**doraval** (*dor-uh-val*) = **Doraemon** + **eval**. The `dora` alias is the same CLI.

Context you cannot rely on wastes a million tokens. doraval makes the first attempt succeed — across Claude, Cursor, Codex, Copilot, and Grok.

## First win

```bash
npx @hacksmith/doraval
# after install: dora
```

Zero config. No API key. Bare `dora` scans the repo: which agents are configured, every skill's health, cross-agent contradictions, and the exact next command to run.

```text
$ dora

  doraval v0.6.x
  Read-only scan of agent context — no writes, no LLM.

  Agent surfaces
    ✓ claude    CLAUDE.md  .claude/skills
    ⚠ cursor    not configured

  Health
    ✓ .claude/skills/review    valid
    ✗ .claude/skills/deploy    Missing "description"

  Next
    1. dora fix .claude/skills/deploy
    2. dora review --all
```

## The loop

```bash
dora                                          # scan
dora new skill --for claude --name review-pr --description "Reviews PRs" --yes
dora review .                                 # structure + heuristics (+ LLM when available)
dora fix .                                    # diffs first; asks before writing
dora memory add "Run tests before shipping skill changes" --weight 8
```

<CardGroup cols={3}>
  <Card title="Get started" href="/get-started/" icon="rocket">
    Install and run your first scan in under two minutes.
  </Card>
  <Card title="Commands" href="/commands/scan/" icon="terminal">
    Full reference for scan, review, fix, new, memory, and more.
  </Card>
  <Card title="For coding agents" href="/for-agents/" icon="bot">
    JSON mode, CI, exit codes, and machine-readable docs.
  </Card>
</CardGroup>

## Next

- [Installation](/get-started/installation/) — npm, Homebrew, Bun
- [Quickstart](/get-started/quickstart/) — scan → new → review → fix → memory
- [Quickstart](/commands/scan/) — full tutorial

---

# dora bump
Source: https://doraval.dev/commands/bump

```bash
dora bump
dora bump patch
dora bump minor ./plugins
dora bump 1.2.3 --yes
dora bump --only plugin
```

| Argument / flag | Meaning |
| --- | --- |
| `TYPE` | `patch` \| `minor` \| `major` \| exact `x.y.z`. Omit on a TTY for interactive pick |
| `PATH` | Directory to scan (default cwd). Single plugin or marketplace root |
| `--only all\|plugin\|marketplace` | Scope (default `all`) |
| `--yes` | Skip confirmation in interactive mode |

Supports Claude, Codex, Cursor, and Copilot packaging layouts.

## Related

- [dora new](/commands/new/)
- [dora providers](/commands/providers/)

---

# dora completion
Source: https://doraval.dev/commands/completion

```bash
dora completion bash
dora completion zsh
dora completion fish
```

Install the printed script into your shell’s completion directory (or eval it from your rc file).

## Related

- [Getting started](/get-started/)

---

# dora config
Source: https://doraval.dev/commands/config

```bash
dora config
dora config get
dora config get eval.model
dora config set eval.model gpt-4.1
```

Common keys: `eval.model`, `eval.provider`, `eval.judge`, `agent.command`. Run `dora config get` for the full table.

Bare `dora config` opens the interactive surface when on a TTY; use `get` / `set` for scripts.

## Related

- [dora review](/commands/review/) (judge / model config)
- [For coding agents](/for-agents/)

---

# dora fix
Source: https://doraval.dev/commands/fix

```bash
dora fix [PATH]
```

Applies mechanical fixes from scan/review findings. Always shows diffs; asks before writing unless `--yes`.

## Options

| Flag | Meaning |
| --- | --- |
| `--yes` | Pre-approve all fixes (agents / CI) |
| `--dry-run` | Show diffs; write nothing |
| `--brief` | Emit an agent-ready prompt for judgment fixes |
| `--format table\|json` | Output format |
| `--ci` | Machine mode (implies JSON) |
| `--cwd <path>` | Working directory override |

## Examples

```bash
dora fix
dora fix ./skills/deploy
dora fix --dry-run
dora fix --yes
dora fix --brief
```

## Related

- [dora review](/commands/review/)
- [dora fix](/commands/fix/)

---

# dora memory
Source: https://doraval.dev/commands/memory

```bash
dora memory <subcommand>
```

Local project principles that stick across sessions. Review treats them as a rubric. Optional promote writes hard rules into `AGENTS.md`. Optional sync backs up to a private git repo.

> Memory replaced the old journal command. The first `dora memory` run migrates legacy journal entries once when present.

## add

```bash
dora memory add "Never use default exports" --weight 8
dora memory add "Run bun test before commit" --tags style,ci --body "CI fails otherwise"
dora memory add "Prefer named exports" --global
```

| Flag | Meaning |
| --- | --- |
| `--weight 1–10` | Default `5`. **w8** hard rule · **w5** default · **w3** soft preference |
| `--tags <csv>` | Comma-separated tags |
| `--global` | Global memory (shared across projects) |
| `--body <text>` | Rationale / detail |

Title is required, short imperative (≤ 80 chars).

## list

```bash
dora memory list
dora memory list --format json
```

## context

Output principles for hooks, `CLAUDE.md`, or piping.

```bash
dora memory context
dora memory context --full
dora memory context --json --quiet
dora memory context --append-to CLAUDE.md
```

| Flag | Meaning |
| --- | --- |
| `--full` | Full principle body |
| `--json` | Claude SessionStart hook JSON (`hookSpecificOutput.additionalContext`) |
| `--quiet` | Omit plain text when empty (still emits JSON with `--json`) |
| `--append-to <file>` | Append/update a managed section in that file |

## promote

```bash
dora memory promote
dora memory promote --weight 7 --dry-run
dora memory promote --yes
```

Writes high-weight principles into `AGENTS.md` (diff + confirm unless `--yes`).

## stash / restore

Copy gitignored or untracked files into project memory (survives a clean clone), then restore later.

```bash
dora memory stash path/to/notes.md
dora memory restore path/to/notes.md
dora memory restore --dry-run
```

## sync

```bash
dora memory sync
dora memory sync --repo owner/dora-memory
```

First run creates/clones the private backup repo; later runs commit + pull --rebase + push.

## Related

- [Memory concept](/concepts/memory/)
- [dora review](/commands/review/)
- [dora memory](/commands/memory/)

---

# dora new
Source: https://doraval.dev/commands/new

```bash
dora new [TYPE] [NAME]
```

| Type | Meaning |
| --- | --- |
| `skill` | Reusable `SKILL.md` |
| `rule` | Always-on convention |
| `agent` | Subagent role |
| `plugin` | Package to ship |

## Options

| Flag | Meaning |
| --- | --- |
| `-f, --for <agent>` | `claude` \| `cursor` \| `codex` \| `copilot` |
| `-d, --description <text>` | Short description |
| `--intent <intent>` | `self` \| `self-later` \| `distribute` |
| `--native` | Scaffold the agent's native format (not plugin packaging) |
| `-y, --yes` | Skip interactive prompts |
| `--format` / `--ci` / `--cwd` | Shared global flags |

## Examples

```bash
dora new
dora new skill --for claude --name review-pr --description "Reviews PRs" --yes
dora new plugin --for claude --name my-plugin --yes
```

Prefer `dora new --for <agent>`. Advanced provider groups (`dora claude new`, `dora codex new`, …) remain for packaging-specific workflows.

## Related

- [dora new](/commands/new/)
- [dora providers](/commands/providers/)

---

# dora providers
Source: https://doraval.dev/commands/providers

```bash
dora providers
dora providers --json
```

Shows which agents doraval understands and how packaging / keyword discovery works for each.

Prefer scaffolding with `dora new --for <agent>` rather than memorizing every layout.

## Related

- [dora new](/commands/new/)

---

# dora reconcile
Source: https://doraval.dev/commands/reconcile

```bash
dora reconcile
dora reconcile --dry-run
dora reconcile --apply
dora reconcile --apply --yes
```

| Flag | Meaning |
| --- | --- |
| `--apply` | Non-interactive: take each recommended resolution |
| `--dry-run` | List contradictions and planned edits; write nothing |
| `--yes` | Skip final confirmation when applying |
| `--format` / `--cwd` | Shared flags |

Interactive by default: review each contradiction and choose how to resolve it.

## Related

- [dora scan](/commands/scan/)
- [dora memory](/commands/memory/)

---

# dora review
Source: https://doraval.dev/commands/review

```bash
dora review [PATH]
```

`PATH` is a skill dir or project root (default `.`).

## Tiers

1. **Structure** — parseable manifests, required fields, path consistency  
2. **Heuristics** — dead refs, duplicates, keyword-level memory violations  
3. **LLM** — judgment when a judge model is configured  
4. **Sessions** — real-session adherence (adapters land over time)

Tiers that cannot run are skipped unless required by a flag.

## Options

| Flag | Meaning |
| --- | --- |
| `--quick` | Tiers 1–2 only (no LLM) |
| `--deep` | Require LLM tier; exit `2` if no judge |
| `--sessions` | Require session tier (planned; currently a no-op when adapters missing) |
| `--all` | Review every artifact under the path |
| `--fail-on error\|warning` | Exit `1` trigger (default `error`) |
| `--format table\|json` | Output format |
| `--ci` | Machine mode (implies JSON) |
| `--cwd <path>` | Working directory override |

## Examples

```bash
dora review ./skills/deploy
dora review --quick .
dora review --deep ./skills/deploy
dora review --all --quick --ci
dora review CLAUDE.md
```

## Related

- [Review tiers](/concepts/review-tiers/)
- [dora fix](/commands/fix/)
- [Memory](/concepts/memory/)
- [scenarios.yaml](/reference/scenarios/)

---

# dora scan
Source: https://doraval.dev/commands/scan

```bash
dora
dora scan
```

Default entrypoint. Read-only diagnosis of agent context in the current (or `--cwd`) directory.

## Options

| Flag | Meaning |
| --- | --- |
| `--format table\|json` | Human table or machine JSON (default `table`) |
| `--ci` | Machine mode (implies `--format json`) |
| `--cwd <path>` | Directory to scan |

## Exit codes

| Code | Meaning |
| --- | --- |
| `0` | Clean (or informational only) |
| `1` | Issues found |
| `2` | Could not run |

## Examples

```bash
dora --format json | jq '.summary'
dora --ci --cwd "$GITHUB_WORKSPACE"
```

## Related

- [dora scan](/commands/scan/)
- [dora review](/commands/review/)
- [dora fix](/commands/fix/)
- [For coding agents](/for-agents/)

---

# dora sessions
Source: https://doraval.dev/commands/sessions

```bash
dora sessions
dora sessions --agent claude --limit 20
dora sessions show <id>
```

Read-only. Claude Code and Grok today; Codex / Copilot / Cursor adapters planned.

## Options (list)

| Flag | Meaning |
| --- | --- |
| `--agent <name>` | Filter (`claude`, `grok`) |
| `--limit <n>` | Max sessions per agent (default `10`) |
| `--format` / `--ci` | Machine output |

## show

Timeline: user turns, tool calls (names + args summary), skills invoked, token totals when available.

```bash
dora sessions show <sessionId>
```

## Related

- [dora review](/commands/review/) (session tier, as adapters mature)
- [For coding agents](/for-agents/)

---

# dora update
Source: https://doraval.dev/commands/update

```bash
dora update
dora update --check
dora update --yes
dora update --via npm
```

| Flag | Meaning |
| --- | --- |
| `--check` | Only check; do not install |
| `--yes` | Skip confirmation |
| `--via homebrew\|npm\|bun` | Force install method (scripts / CI) |

## Related

- [Installation](/get-started/installation/)

---

# Memory
Source: https://doraval.dev/concepts/memory

**Memory** is doraval’s store of project (and optional global) principles. It is the only path for durable decisions; the old journal command is gone.

## Why it exists

Sessions forget. README bullets rot. Memory keeps short, weighted principles where:

1. Humans can list and edit them (`dora memory list` / `add`)
2. Review can enforce them as a rubric
3. Promote can write hard rules into `AGENTS.md`
4. Hooks can inject them at session start

## Weights

| Weight | Role |
| --- | --- |
| **8–10** | Hard rule — strong review signal; default promote threshold is 7+ |
| **5** | Default preference |
| **1–3** | Soft preference |

## Lifecycle

```bash
dora memory add "…" --weight 8
dora memory list
dora review .                    # principles applied as rubric
dora memory promote              # hard rules → AGENTS.md
dora memory context --json       # SessionStart hook payload
dora memory sync                 # optional private git backup
```

## Stash

Gitignored research files can live in memory and restore after a clean clone:

```bash
dora memory stash path/to/plan.md
dora memory restore path/to/plan.md
```

## Migration note

If you previously used journal, the first `dora memory` command migrates valid entries automatically. Update any SessionStart hooks to:

```bash
dora memory context --json
```

## Related

- [dora memory](/commands/memory/)
- [dora review](/commands/review/)
- [For coding agents](/for-agents/)

---

# Review tiers
Source: https://doraval.dev/concepts/review-tiers

`dora review` runs every tier that can run. Flags force or skip tiers when you need a fixed bar (CI vs deep judgment).

## Tier 1 — Structure

Parseable files, required fields, name/dir consistency, basic schema. Fast. Always available. `--quick` includes this tier.

## Tier 2 — Heuristics

Dead file references, duplicate rules, keyword-level violations of [memory](/concepts/memory/) principles, packaging smells. Still no network / LLM. Included in `--quick`.

## Tier 3 — LLM

Judgment when a judge model is configured (`dora config` / env). Vague descriptions, missing guardrails, scenario coverage against [scenarios.yaml](/reference/scenarios/).

- Default: run if available, skip if not  
- `--deep`: **require** this tier; exit `2` if no judge  

## Tier 4 — Sessions

Compare skill intent to real agent sessions when adapters exist. Session adapters ship over time; `--sessions` will require this tier when ready.

## Choosing a mode

| Goal | Command |
| --- | --- |
| CI structural gate | `dora review --all --quick --ci` |
| Local full pass | `dora review .` |
| Must have LLM | `dora review --deep <path>` |

## Related

- [dora review](/commands/review/)
- [dora fix](/commands/fix/)
- [scenarios.yaml](/reference/scenarios/)

---

# Fix
Source: https://doraval.dev/get-started/quickstart/fix

When the scan (or review) lists auto-fixable issues:

```bash
dora fix
dora fix ./skills/deploy
dora fix --dry-run          # show diffs, write nothing
dora fix --yes              # pre-approve (agents / CI)
dora fix --brief            # agent-ready prompt for judgment items
```

**Mechanical** fixes (rename fields, missing frontmatter, path mismatches) show as unified diffs and apply per-file on approval.

**Judgment** items (vague descriptions, missing guardrails) are not auto-written. Use `--brief` to copy a prompt into your coding agent, then re-run `dora review`.

Full reference: [dora fix](/commands/fix/).

**Next:** [Review](/commands/review/) for multi-tier quality.

---

# Memory
Source: https://doraval.dev/get-started/quickstart/memory

```bash
dora memory add "Never use default exports" --weight 8
dora memory list
dora memory promote                 # hard rules → AGENTS.md
dora memory context --json --quiet  # SessionStart hooks
```

Weights: **w8** hard rule · **w5** default · **w3** soft preference.

Review uses principles as a rubric. Promote writes high-weight items into `AGENTS.md` (diff + confirm).

Full reference: [dora memory](/commands/memory/) · [Memory concept](/concepts/memory/).

**Next:** [Commands](/commands/scan/) for the full atlas, or [For coding agents](/for-agents/).

---

# New
Source: https://doraval.dev/get-started/quickstart/new

```bash
dora new
dora new skill --for claude --name review-pr --description "Reviews PRs" --yes
dora new rule --for cursor --name no-default-exports --yes
dora new plugin --for claude --name my-plugin --yes
```

| Type | What it is |
| --- | --- |
| `skill` | Reusable `SKILL.md` |
| `rule` | Always-on convention |
| `agent` | Subagent role |
| `plugin` | Package to ship |

Prefer `dora new --for <agent>`. Advanced packaging groups (`dora claude new`, …) still exist when you need provider-specific trees.

Full reference: [dora new](/commands/new/).

**Next:** [Memory](/commands/memory/) for principles that stick.

---

# Review
Source: https://doraval.dev/get-started/quickstart/review

```bash
dora review .
dora review ./skills/deploy
dora review --all
dora review --quick .           # structure + heuristics only
dora review --deep ./skills/foo # require LLM tier (exit 2 if no judge)
dora review --all --quick --ci  # CI gate
```

Tiers that can run do run; LLM and sessions auto-skip when unavailable (unless `--deep` / `--sessions` require them).

Project principles from [memory](/concepts/memory/) apply automatically as a rubric.

Full reference: [dora review](/commands/review/) · [Review tiers](/concepts/review-tiers/).

**Next:** [Scaffold](/commands/new/) new context with `dora new`.

---

# Scan
Source: https://doraval.dev/get-started/quickstart/scan

Your first command is the default:

```bash
dora
# same as:
dora scan
```

Read-only. No writes, no LLM. Output covers:

1. **Agent surfaces** — which agents look configured (CLAUDE.md, skills dirs, …)
2. **Health** — skill/manifest issues with short reasons
3. **Next** — numbered commands to run next (`dora fix …`, `dora review …`)

Useful flags:

```bash
dora --format json
dora --ci                 # implies JSON
dora --cwd /path/to/repo
```

Full reference: [dora scan](/commands/scan/).

**Next:** [Fix](/commands/fix/) mechanical issues the scan found.

---

# For coding agents
Source: https://doraval.dev/for-agents

doraval is built so coding agents and CI can drive it without a human reading tables.

## Machine output

```bash
dora --format json
dora --ci                    # implies --format json
dora review --all --quick --ci
dora --capabilities          # JSON manifest of commands (not a human command)
```

Global flags:

| Flag | Meaning |
| --- | --- |
| `--format table\|json` | Output format |
| `--ci` | Machine mode (JSON) |
| `--cwd <path>` | Working directory |
| `--capabilities` | Emit command capability manifest |

## Exit codes

| Code | Meaning |
| --- | --- |
| `0` | Clean |
| `1` | Issues found |
| `2` | Could not run (e.g. required LLM tier unavailable with `--deep`) |

## CI example

```bash
npx @hacksmith/doraval review --all --quick --ci
dora --format json | jq '.summary'
```

## Agent-ready fixes

```bash
dora fix --yes               # pre-approve mechanical fixes
dora fix --brief             # clipboard/prompt for judgment items
dora fix --dry-run
```

## Machine-readable docs

This site ships agent surfaces by default:

| URL | Content |
| --- | --- |
| [https://doraval.dev/llms.txt](https://doraval.dev/llms.txt) | Compact index of every page |
| [https://doraval.dev/llms-full.txt](https://doraval.dev/llms-full.txt) | Full corpus Markdown |
| Any page + `.md` | Raw Markdown for that page (e.g. `/commands/review.md`) |

## Hooks

Inject project principles into Claude SessionStart:

```bash
dora memory context --json --quiet
```

## Related

- [dora scan](/commands/scan/)
- [dora review](/commands/review/)
- [dora memory](/commands/memory/)

---

# Getting started
Source: https://doraval.dev/get-started

## Who this is for

Anyone **scaling AI context** for coding agents — skills, plugins, decisions, and the glue between them — for **yourself**, **your team**, or **your community**.

## The problem

Context you cannot rely on wastes a million tokens. Broken manifests, skills that never trigger, decisions that evaporate between sessions, “works on my Claude” that fails on Cursor, Codex, or the tenth engineer you onboard.

## The win

Left-shift success. **Scan → review → fix → remember**, so the first attempt works across agents.

| Job | Command |
| --- | --- |
| See what's broken | `dora` / `dora scan` |
| Grade quality | `dora review` |
| Apply mechanical fixes | `dora fix` |
| Scaffold context | `dora new --for <agent>` |
| Principles that stick | `dora memory` |

## Paths

1. [Install](/get-started/installation/)
2. [Quickstart](/get-started/quickstart/) — first win under two minutes
3. [Quickstart](/commands/scan/) — full tutorial

---

# Installation
Source: https://doraval.dev/get-started/installation

## No install

```bash
npx @hacksmith/doraval
```

## npm

```bash
npm install -g @hacksmith/doraval
```

npm ships a prebuilt binary per platform (macOS arm64/x64, Linux x64/arm64, Windows x64). Node ≥ 14.18.

## Homebrew (macOS)

```bash
brew tap saif-shines/tap && brew trust saif-shines/tap && brew install doraval
```

## Bun

```bash
bunx @hacksmith/doraval
bun add -g @hacksmith/doraval
```

## Alpine / musl

Platform binaries target glibc. On Alpine/musl, run from source with Bun:

```bash
bunx @hacksmith/doraval
```

## Verify

```bash
doraval --version
# same binary:
dora --version
```

## Next

[Quickstart](/get-started/quickstart/) — run your first scan.

---

# Quickstart
Source: https://doraval.dev/get-started/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.

Assumes doraval is installed. If not, see [Installation](/get-started/installation/).

## 1. Scan what you already have

```bash
dora
```

Bare `dora` lists agent surfaces, skill health, contradictions, and numbered next actions. No writes, no API key.

## 2. Scaffold context

```bash
# Interactive: type → agent → name
dora new

# Non-interactive
dora new skill --for claude --name my-context --description "Team context" --yes
dora new plugin --for claude --name my-plugin --yes
```

Prefer `dora new --for <agent>`. Thin wrappers (`dora claude new`, …) still work for advanced packaging.

## 3. Review and fix

```bash
dora review .
dora fix .                 # mechanical fixes; always shows a diff
```

CI-friendly structural gate (no LLM):

```bash
dora review --all --quick --ci
```

## 4. Record principles that stick

```bash
dora memory add "Use dora review before shipping any skill change" --weight 8
dora memory list
dora memory promote        # optional: write hard rules into AGENTS.md
dora memory sync           # optional: private git backup
```

For Claude SessionStart hooks:

```bash
dora memory context --json --quiet
```

## Next

- [dora scan](/commands/scan/)
- [For coding agents](/for-agents/) — JSON, CI, exit codes
- [Memory concept](/concepts/memory/)

---

# scenarios.yaml
Source: https://doraval.dev/reference/scenarios

Place an optional `scenarios.yaml` next to a skill. Missing file is fine (empty coverage set).

## Shape

YAML **array** of objects:

| Field | Required | Meaning |
| --- | --- | --- |
| `when` | yes | User request / situation |
| `expect` | yes | Expected agent behavior |
| `must_not` | no | Forbidden behavior |

```yaml
- when: "User asks to deploy without tests"
  expect: "Refuse and require the test suite to pass first"
  must_not: "Run deploy immediately"

- when: "User asks for a PR review"
  expect: "Load the review skill and comment on risk first"
```

## How doraval uses it

During LLM review (tier 3), scenarios become a coverage checklist: for each item, would this skill handle `when` correctly per `expect` / `must_not`? Only **uncovered** scenarios become findings.

Malformed YAML or invalid objects surface as tier-1 structure findings, not a crash.

## Related

- [Review tiers](/concepts/review-tiers/)
- [dora review](/commands/review/)