cross-review

A code-review workflow that asks independent Claude and GPT-5.5 reviewers to inspect the same changes, compare their findings, and produce one prioritized fix list.

In plain words
What is it for?
Reviewing uncommitted changes, selected files, commit ranges, or described areas of a codebase, with optional quick or reconciliation modes.
Why use it?
It reduces the chance that a single review misses a serious bug or raises an issue without enough evidence. Cross-checking helps separate material problems from minor style concerns.

Skill for Claude CodeCodex

Part of the cross-review plugin — 1 skill, 2 agents shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/kelp/agent-plugins/cross-review
Any agent
npx skills add kelp/agent-plugins --skill cross-review
Clone the repo
git clone --depth 1 https://github.com/kelp/agent-plugins

Made for: Claude Code, Codex.

Or install cross-review, the plugin that ships this one along with the rest of its 1 skill, 2 agents.

Per session 95 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 4,631 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00095 $0.04631
Opus 5 $0.00048 $0.02315
Sonnet 5 $0.00019 $0.00926
Haiku 4.5 $0.00010 $0.00463

Measured 2d ago against content hash 32188677d671, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

cross-review scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Recursive force deletemediumDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

trap 'rm -rf "$cr_tmpdir"' EXIT

Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.

plugins/cross-review/skills/cross-review/SKILL.md · 513 lines

How it starts

The opening of the file, as written. The whole thing — 513 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/cross-review

Independent reviews from two models, cross-validated, merged into a single prioritized fix list.

Input

Scope: $ARGUMENTS

Default (no arguments): all uncommitted changes, equivalent to git diff HEAD.

Accepted scope forms:

  • File path: src/server.zig
  • Git range: last 2 commits, HEAD~3..HEAD
  • Any freeform description — translate it to the appropriate git diff or file list.

Flags:

  • --quick: skip cross-validation; merge findings from both models without checking each other's work.
  • --reconcile: after cross-validation, ask each model to reconsider its disputed findings in light of the validator's reasoning. Incompatible with --quick. Adds one round of reconciliation per disputed finding.
  • --model <name>: run the Claude-side reviewer and validator agents on a specific model (e.g. opus) instead of the session model. Does NOT affect the codex/GPT side — that model is set by your codex install. With no flag, the agents inherit the session model.

The Rule

You are a PURE DISPATCHER. You NEVER edit source files. Your job is to package context, dispatch reviewer and validator agents, shell out to codex, normalize findings, and merge results. Never fix an issue you find, never inline a fix in the review output, never skip a pipeline stage ("this finding is obvious", "codex is unavailable, I'll skip the merge"), and never deduplicate raw prose without normalizing first. If you catch yourself about to use Write or Edit on a source file, STOP — format the finding and move on.


Pipeline

Step 1: Determine Scope and Package Context

Parse $ARGUMENTS:

  • Strip --quick and --reconcile flags if present; record each one as a boolean.
  • Strip --model <name> if present; record the model name. When set, pass model: <name> to every cross-review:reviewer and cross-review:validator Agent dispatch below. When absent, omit model: so those agents inherit the session model. This flag never affects the codex/GPT side.
  • If BOTH --quick and --reconcile are set, stop immediately and report an error — they are incompatible (quick mode skips cross-validation, so there is nothing to reconcile).
  • Remaining text is the scope. If empty, use git diff HEAD.
  • Translate freeform scope to git commands or file paths as needed.

Read the full file on GitHub · 513 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 513 lines · 95 tokens per session scan B 32188677d671

Subscribe to this mod's changes

cross-review is a skill published in the GitHub repository kelp/agent-plugins (2 stars, last pushed 10d ago), licensed MIT. It adds 95 tokens to every session and 4,631 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it B with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

strategy-red-team

Red-team a PRD, roadmap, or strategy by attacking its load-bearing assumptions before reality does. Steelmans then attacks each claim, ranks failure modes by impact × likelihood × cheapness-to-test, and returns the cheapest test and kill criteria for each. Use when stress-testing a plan, pressure-testing a strategy…

phuryn/pm-skills · 78 tokens

retro

Facilitate a structured sprint retrospective — what went well, what didn't, and prioritized action items with owners and deadlines. Use when running a retrospective, reflecting on a sprint, creating action items from team feedback, or learning how to run effective retros.

phuryn/pm-skills · 52 tokens

release-notes

Generate user-facing release notes from tickets, PRDs, or changelogs. Creates clear, engaging summaries organized by category (new features, improvements, fixes). Use when writing release notes, creating changelogs, announcing product updates, or summarizing what shipped.

phuryn/pm-skills · 57 tokens

competitive-battlecard

Create sales-ready competitive battlecards comparing your product against a specific competitor — positioning, feature comparison, objection handling, and win/loss patterns. Use when preparing sales teams, creating competitive materials, or responding to 'why not competitor X?'.

phuryn/pm-skills · 52 tokens

stakeholder-map

Build a stakeholder map using a power/interest grid, identify communication strategies per quadrant, and generate a communication plan. Use when managing stakeholders, preparing for a launch, aligning cross-functional teams, or planning stakeholder engagement.

phuryn/pm-skills · 47 tokens

ms

Skill "ms" from boshu2/agentops, covering ms — metaskill search/load engine, constraints, quick start, consume — mcp-primary (mcpms) and write / admin — cli-only (verified landing in the live db).

boshu2/agentops · 65 tokens