skills.rest
k0d3x8its
k0d3x8itsCOMMUNITY
Explore all skills by this author →
Website𝕏@k0d3x8itsGitHub

code-crit

Reviews code diffs using parallel persona agents with severity-grouped findings reports.

3|Updated Oct 28, 2020
Search Tags:#git#diff analysis#code review#pull request#multi-agent#security review#spec compliance
One-click install
npx skills add https://github.com/k0d3x8its/dotfiles --skill code-crit-k0d3x8its
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: code-crit
Source: https://github.com/k0d3x8its/dotfiles/tree/main/claude/.claude/skills/code-crit
Command: npx skills add https://github.com/k0d3x8its/dotfiles --skill code-crit-k0d3x8its

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Reviewing large diffs, branches, or pull requests by hand is slow and inconsistent, and single-pass reviews miss issues across security, correctness, testing, and spec compliance. This Skill runs a structured multi-persona code review that produces a deduplicated, severity-ranked report with verified/unverified confidence labels. ## Core Features & Use Cases - Parallel persona review: Dispatches up to 12 specialized reviewer personas (correctness, security, spec-compliance, adversarial, maintainability, testing, and more) against a diff, with always-on and conditionally-triggered roles. - Two review modes: A fast default mode that batches Sonnet-tier personas into one mega-spawn, and a thorough mode that fully isolates every persona for maximum coverage. - Stage-2 synthesis and reporting: An Opus advisor pass deduplicates and reranks findings, then emits a Spec compliance section plus a severity-grouped findings table with confidence and routing labels. - Use Case: Run a thorough review of a pre-release feature branch to catch security regressions, spec mismatches, and untested edge cases before merging. ## Quick Start Ask the AI to run code-crit on the current branch diff, optionally adding the word thorough for a fully isolated deep review.

Frequently Asked Questions about code-crit

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

FAQPage Schema
How do I review a pull request with multiple AI personas?▼

Invoke code-crit with a PR number or branch name and it resolves the diff, dispatches up to 12 specialized reviewer personas in parallel, and synthesizes their findings into one severity-grouped report. Each finding carries a verified or unverified confidence label.

What is the difference between fast and thorough code review modes?▼

Fast mode is the default and batches all matched Sonnet-tier personas into one shared mega-spawn to save tokens. Thorough mode runs every persona as its own fully isolated spawn, preventing one persona's blind spot from suppressing another's findings, at roughly 12 spawns instead of 5.

How does code-crit differ from the built-in /code-review command?▼

code-crit is a local skill using parallel persona agents with binary confidence labels and a Spec-plus-severity report. The built-in /code-review and /code-review ultra are separate cloud-based commands, and the skill explicitly warns not to confuse the two.

Can code-crit handle very large diffs or feature branches?▼

Yes, large diffs are batched by diff line count with a target ceiling of roughly 1500-1600 lines per batch, grouped by subsystem. Batches are tracked in a tracker file and findings file so multi-session reviews can resume safely.

What happens when a review persona fails during a run?▼

In fast mode, if the Sonnet-tier mega-spawn fails to produce parseable JSON, the run proceeds with only the Opus-frontline findings and the report opens with an explicit failure marker. The skill never retries silently or fabricates missing findings.

Related Skills

Explore More Software Engineering Skills→
reactivereactive
2.0k

pr

Creates GitHub pull requests from working changes with generated titles and descriptions.

Official
Intermediate
kosakokosako

personal-repo-audit

Audit a repository across multiple dimensions and aggregate severity-ranked findings with file locations.

Community
Intermediate
gmackiegmackie

interrogate

Spawns multiple LLM reviewers to adversarially review code changes and synthesizes a verdict.

Community
Advanced
AbildtoftAbildtoft
2

kramme:resolve-review:team

Divide code review findings among specialized agent teams for parallel resolution.

Community
Advanced