Generate standardized reviewer prompts for multiple review passes against specs.
npx skills add https://github.com/ramboz/jig --skill independent-review
Please help me install this Agent Skill. Skill: independent-review Source: https://github.com/ramboz/jig/tree/main/skills/independent-review Command: npx skills add https://github.com/ramboz/jig --skill independent-review
This skill creates a deterministic, frontmatter-driven reviewer prompt that lets Claude act as a read-only reviewer, evaluating deliverables against their spec without exposing internal implementation conversations.
Run python3 review.py implementation <spec.md> <slice-fragment> <deliverable-path>... to generate the prompt and feed it to Claude's Task tool.
High-intent search queries and answers about installing and using this skill.
Standardized code review prompts evaluate deliverables against specs by acting as a read-only reviewer, limiting access to spec and deliverable inputs while producing a normalized verdict envelope. This prevents exposing internal implementation conversations during the review process.
A read-only reviewer workflow for architecture review evaluates deliverables against target specs by limiting access to slice metadata and deliverable paths, ensuring the reviewer sees only the output while producing a normalized verdict envelope for downstream tools.
Generating multiple review passes involves constructing standardized prompts for implementation, pr-review, arch-review, and reconciliation by running a script with the target spec, slice fragment, and deliverable paths to produce a single normalized verdict envelope.
Custom installed skills can override baseline reviewer prompts if they are installed on disk, allowing richer skills to accommodate multiple modes like implementation, reconciliation, arch-review, design-review, and code-health.
Parsing a normalized verdict envelope after a code review involves feeding the generated reviewer prompt to Claude's Task tool, which outputs a standardized envelope that downstream tools can parse to determine compliance and craft quality.
Reviewer prompts need slice metadata and a target spec to deterministically evaluate work against specifications, ensuring the reviewer operates in a read-only context and produces a normalized verdict envelope without seeing implementation details.
rambozAudit a jig spec against related artifacts and emit a markdown findings report.
Davie521Apply dual independent reviews to verify generated outputs before shipping.
samjhechtGenerate task-level implementation plans with exact file paths and code samples.
miles990Dispatch a code-reviewer subagent to review changes before merging.
