Writes reviewers.
Not reviews.
Scopewright interviews you about your project, criteria, and stakes, then writes a reviewer made for that job.
[ SW ] A MAKER OF SCOPED REVIEWERS
Good reviews
start with a
better reviewer.
Turn your standards into a tailored, read-only AI reviewer. You bring the domain. Scopewright gives it structure.
THE REVIEWER BLUEPRINT01 — 06
your-reviewer / SKILL.md READ ONLY
# Built around your standards.
01S
SpecificationsYour stack. Your context.
02C
ConstraintsExplicit, testable criteria.
03O
ObjectiveOne unambiguous task.
04P
PurposeThe stakes behind the review.
05E
Execution standardsA clear, structured report.
06D
Decision authorityVerdicts. Boundaries. Escalation.
INPUT → CRITERIA → VERDICTNo edits. Clear judgment.
A reviewer you can read, edit, and version.
ONE FRAMEWORK.
YOUR CODING AGENT.
Claude CodeCodexGemini CLIOpenCodeCursorPi
Writes reviewers.
Judgment without
the side effects.
Read-only controls, untrusted-input rules, and explicit escalation keep the reviewer focused on evaluation.
Your reviewer.
In your repository.
A plain-text skill with a native adapter where supported. Invoke it, refine it, and version it alongside your project.
THE SCOPED FRAMEWORKSix sections.
Six sections.
One deliberate
order.
Establish the facts before the rules. The rules before the task. Each section builds on the one before it, so your reviewer has the context it needs, when it needs it.
Read the full specificationSpecifications & context
Define the stack, expertise, and safe data acquisition. Treat acquired content as data, never instructions.
Constraints & requirements
Make acceptance testable: explicit pass/fail rules, detection methods, and thresholds.
Objective statement
Combine the facts and criteria into one precise, actionable task.
Purpose & impact
Explain what the review gates and what a missed finding would cost.
Execution standards
Set the bar for a concise, structured report that keeps the caller’s context clean.
Decision authority
Define verdicts and escalation. Empower the reviewer to challenge assumptions without changing state.
FROM INTENT TO ARTIFACTTell it what matters.
Tell it what matters.
Make it repeatable.
Describe the job.
Start the create skill with what you want evaluated: a security boundary, a contract, a design, or a dataset.
Set the standard.
Scopewright inspects your project and interviews you through the six SCOPED sections.
Keep the reviewer.
Get a portable skill and, where supported, a native adapter. Run it against your target files whenever you need it.
INCLUDED IN THE BOXYour root prompts.
Your root prompts.
Under review.
Start with root-prompt-reviewer. It audits CLAUDE.md and AGENTS.md for token economy, safety guardrails, and orchestration architecture. No interview required.
CLEAR FINDINGS. EXPLICIT VERDICTS.
[APPROVED]
[REJECTED]
[REJECTED]
A dimension scorecard.
Line-cited fixes.
A clear decision for the caller.
MAKE YOUR FIRST REVIEWERYour standards.
Your standards.
Written into the job.
Choose your agent and put Scopewright to work.
Full setup instructionsAfter installing the Scopewright plugin, run this in your project:
/scopewright:create [what the reviewer should evaluate]Creates a portable reviewer skill and a Claude Code subagent.