plan-critic

Produces structured JSON critiques of implementation plans against a fixed schema.

9|Updated Jun 10, 2026
One-click install
npx skills add https://github.com/eventbalancer/agent-quorum --skill plan-critic-eventbalancer
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: plan-critic
Source: https://github.com/eventbalancer/agent-quorum/tree/main/skills/plan-critic
Command: npx skills add https://github.com/eventbalancer/agent-quorum --skill plan-critic-eventbalancer

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Reviewing implementation plans by hand is inconsistent: reviewers miss risk domains, repeat already-resolved findings, and return free-form feedback that automated pipelines cannot validate. This Skill turns plan review into a deterministic, schema-validated JSON critique that an orchestration loop can consume directly. ## Core Features & Use Cases - Schema-conformant critique output: Emits JSON only, conforming to critique.schema.json, with domain assessments, issues, boundary challenges, and opportunities. - Eight fixed readiness domains: Classifies correctness, public-compatibility, data-migrations, security, concurrency, cross-repository delivery, operability, and performance with applicability and risk levels. - Bounded issue budget with lineage: Reports at most 8 material blocker/major issues, each with evidence, severity, confidence, and versioned parent references (vN.Cm) to avoid repeating resolved findings. - Use Case: Inside the agent-quorum plan → critique → update loop, the critic reviews plan.v2.md, verifies file references via targeted reads, and returns a validated JSON critique that decides whether the plan needs another revision. ## Quick Start Ask the agent to critique the current implementation plan and return only JSON conforming to critique.schema.json with domain assessments and material issues.

Frequently Asked Questions about plan-critic

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get structured JSON feedback on an implementation plan?

Provide the plan markdown plus the retained run context, and the critic returns JSON only, conforming to critique.schema.json. The output includes a summary, eight domain assessments, material issues, boundary challenges, and opportunities with no prose or markdown fences.

What makes a plan review issue a blocker versus a major?

A blocker means the plan cannot be executed as-is, such as a logical contradiction, violated hard constraint, or data loss risk. A major is a significant gap like a missing test strategy or unverified technical claim that reduces execution confidence but does not break the plan outright.

Can the critic modify code or files while reviewing a plan?

No. The critic is explicitly read-only: it may use targeted Read operations to verify file and API claims in the plan, but it cannot edit files, run Bash, or execute the plan. Its only output is the JSON critique.

Why does a critique get rejected by schema validation?

Common causes include a summary over 700 characters, missing required fields like review or domain_assessments, fewer or more than eight domain assessments, or issue IDs not matching the C1/C2 pattern. Any schema violation aborts the loop iteration.

What are the limitations of this plan critique approach?

The critic reports at most 8 material issues per pass, so an exhausted budget marks scan_complete as false. It also cannot judge effort-estimate accuracy, propose fallback strategies, or expand the frozen scope; those require boundary challenges or remain out of scope.