CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

review-with-dora

Use when reviewing agent context (SKILL.md, plugin, rule, AGENTS.md) with dora before reporting done. Also when ask-dora routes a Dora job (review, fix, unused, scan), or the user asks to review, diagnose, or clean up Claude, Cursor, Codex, Copilot, or Grok context.

68

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

review-with-dora

You run dora. Assess the agent setup in this project and show what to fix. Review is the gate. Done means exit 0. Ship a repeating job as a pocket agent.

If dora is not on PATH, use npx @hacksmith/doraval.

The loop

  1. Review. dora review --quick on a new repo. After an edit, dora review --quick <path>. <path> is the Skill directory, a plugin root, or .. More than 10 artifacts: pass a path or --all. Done: Findings read, exit code read. exit 0 is the only report-done.

  2. Author only when the user asked you to write a Skill, Rule, agent, or plugin. dora skill new / dora rule new / dora agent new / dora plugin new. Done: the files exist.

  3. Mechanical fix.

    dora fix <path> --dry-run
    dora fix <path> --yes

    --yes applies frontmatter, formatting, and missing fields. Done: dry-run is clean, or --yes applied.

  4. Judgment. Run Brief. Then replace. dora fix <path> --brief (or --json). Each item: message, severity, hint, optional code / docUrl. Target is that Authored SKILL.md.

    Replace one item at a time:

    • Name the intended rule in one sentence. Write the smallest wording that states it.
    • Put that wording where the old sentence lived. One home.
    • Stay inside the span the item names.
    • Imperative. One statement of the rule.
    • Skill already states the rule → show the Finding, leave the file.
    • Stay in this Skill's files. A size Finding may move text into this Skill's references/.

    Done: every item replaced or shown. Then step 5.

  5. Re-review. dora review --quick <path> after every edit. Done: exit 0. That is the only report-done.

  6. Delegated Judge only after step 5 is exit 0 and a later Review (no --quick) is delegated. Read JUDGE THIS. Evaluate. Fix Findings. Done: mechanical tiers clean and the rubric applied. Branch on JSON: output.md.

Writes

Pass --yes or --dry-run on fix, conflicts, memory promote, skill remove, skill restore, probe, harness new --accept, harness apply, harness rm, and config set of a secret. --brief is not a write. A missing write flag is exit 2.

Side paths

Verbs: commands.md. Exit codes and JSON: output.md.

dora wire --for claude --run records the request the model received. It is not a Finding. Logs stay in ~/.doraval/wire/.

Recurring use case → dora harness new (starts ask-dora / grilling-for-routine). Then dora harness apply <slug>. dora harness boot <slug> is the apply alias. Then hermes mcp login scalekit. Runtime watch after apply, boot, list, pause, and resume. Also dora harness show <slug>, dora harness models, dora harness logs <slug>, dora harness rm <slug> --yes, dora harness pause, dora harness resume, dora harness list, dora harness open.

MUST report done only on exit 0. MUST NOT skip Brief for Judgment items.

Repository

saif-shines/doraval

Path

skills/review-with-dora/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.