neo-code-review
Skill7 starsChanged 6 months agoMITView on GitHub
Use this skill when the user asks to review or audit source code, a PR, diff, commit, or recent changes for bugs, security, performance, tests, compatibility, or maintainability, duplicated code or logic across files, or hard-coded values. Also use it after an AI agent finishes modifying code to inspect the current uncommitted working tree before handoff and report actionable findings with file and line evidence.
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- generate-skillSkill
- neo-azure-pipelinesSkill
- neo-clarificationSkill
- neo-clean-architectureSkill
- neo-csharp-interface-generatorSkill
- neo-csharpSkill
- neo-dotnet-ef-coreSkill
- neo-dotnet-minimal-apisSkill
- neo-dotnet-mvcSkill
- neo-dotnetSkill
- neo-dotnet-tag-helperSkill
- neo-dotnet-webapiSkill
- neo-harness-projectSkill
- neo-harnessSkill
- neo-javascriptSkill
- neo-opentelemetrySkill
- neo-prSkill
- neo-python-managerSkill
- neo-pythonSkill
- neo-rabbitmqSkill
- neo-rustSkill
- neo-stop-slopSkill
- neo-sub-agentSkill
- neo-swiftSkill
- neo-swift-uiSkill
- neo-typescriptSkill
- neo-vueSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 7 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Follow this read-only workflow.
- Distinctive: line 25
Use the files or snippets supplied by the user as the primary scope.
- Distinctive: line 26
If the user provides no scope and the current directory is a Git repository, run:
- Very specific: line 30
Use --staged for staged-only changes and --commit <range> for a commit range.
- Distinctive: line 31
Read surrounding code, related tests, and existing contracts.
- Distinctive: line 35
Read review-checklist.md before evaluating code.
- Very specific: line 44
If the user supplied only a snippet or specific files without repository context, state in ## Validation that cross-file verification was …
- Distinctive: line 51
Do not report preferences, unsupported speculation, or checklist items that were not triggered by evidence.
- Very specific: line 54
Write in English and preserve file paths and line numbers, such as src/user.ts:42.
- Very specific: line 55
Use this structure. Keep all severity sections and write None when a section has no findings:
- Very specific: line 83
If an automatic review finds 🔴 Critical Issues, cannot establish the review scope, or cannot load the skill, do not declare completion or …
- Distinctive: line 84
After fixing findings, rerun the required checks and review the changes again.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Code Review
Follow this read-only workflow. Report only actionable findings supported by evidence.
Trigger On
- The user requests a code, PR, diff, commit, security, performance, test, or maintainability review.
- The user asks whether code or logic is duplicated across files, or whether hard-coded values should be centralized.
- An AI Agent finishes modifying code, tests, configuration, or scripts before handoff.
Workflow
- Define the review scope
- Use the files or snippets supplied by the user as the primary scope.
- If the user provides no scope and the current directory is a Git repository, run:
uv run skills/neo-code-review/scripts/git-diff-reviewer.py --working-tree - Use
--stagedfor staged-only changes and--commit <range>for a commit range. - Read surrounding code, related tests, and existing contracts. If no reviewable input exists, report it and stop.
- For a working-tree review, include the helper's tracked diff and each non-ignored untracked file as separate inputs.
- Load the review rules
- Read review-checklist.md before evaluating code.
- Identify the actual language, framework, and affected behavior from the change. Do not infer unverified rules from filenames.
- Run the cross-file consistency scan
- Build the comparison set from the review targets and relevant repository source and configuration files. Use
rg --filesand the repository's ignore rules. Exclude dependency, vendor, cache, generated, minified, build, and coverage output unless a file is explicitly in scope. - Inventory changed or reviewed functions, classes, validation and authorization rules, state transitions, request or query construction, error handling, and distinctive literals such as URLs, routes, timeouts, retry counts, thresholds, regexes, feature flags, and error messages.
- Search the comparison set for exact and near-exact matches, then read each match in context. Ignore name and formatting differences when the behavior matches. Treat renamed or slightly rearranged copies as duplicates when their business rules, inputs, side effects, and failure behavior are materially the same.
- Report only meaningful duplication: copied implementation, duplicated business or validation logic, divergent security or state logic, or repeated non-trivial hard-coded configuration. Do not report a match based only on a common keyword, normal one-time literal, test fixture, snapshot, generated file, or vendor code.
- For every cross-file finding, cite both file paths and line numbers. Explain the concrete maintenance, divergence, security, or behavior risk and recommend a specific shared helper, constant, configuration source, or other single source of truth.
- If the user supplied only a snippet or specific files without repository context, state in
## Validationthat cross-file verification was unavailable. Never invent repository matches or line evidence.
- Build the comparison set from the review targets and relevant repository source and configuration files. Use
- Analyze risk
- Prioritize correctness, regressions, security, data consistency, resource lifetime, concurrency, and high-risk test gaps.
🔴 Critical Issuescontain only security vulnerabilities, data loss, incorrect behavior, authorization bypasses, or major regressions.🟡 Suggestionscontain only actionable, non-blocking issues with a clear impact and remediation direction.- Classify ordinary duplicated logic or hard-coded configuration as
🟡 Suggestions. Classify hard-coded secrets or duplication that creates a demonstrable security or behavior defect as🔴 Critical Issues. - Do not report preferences, unsupported speculation, or checklist items that were not triggered by evidence.
- Produce a concise report
- Write in English and preserve file paths and line numbers, such as
src/user.ts:42. Use this structure. Keep all severity sections and write
Nonewhen a section has no findings:## 🔍 Review Summary - **Status**: `🟢 Ready to deliver`, `🟡 Fix recommended`, `🔴 Delivery blocked`, or `⚪ No reviewable changes` - **Scope**: Files or changes that were inspected. ## 🔴 Critical Issues - **Location**: `path:line` - **Problem**: Specific behavioral defect. - **Impact**: Verifiable consequence. - **Fix**: Concrete remediation direction; include code only when it clarifies the fix. ## 🟡 Suggestions - **Location**: `path:line` - **Type**: Duplicated logic, duplicated hard-coded value, or another applicable category. - **Evidence**: The matching location and the relevant behavior or value in each file. - **Recommendation**: Specific, non-blocking improvement. ## 🟢 Strengths - List only strengths directly supported by the change. ## Validation - State the cross-file comparison scope and any unavailable repository context. - List executed tests, builds, and checks. Name any that were not run.
- Write in English and preserve file paths and line numbers, such as
- Apply the delivery gate
- If an automatic review finds
🔴 Critical Issues, cannot establish the review scope, or cannot load the skill, do not declare completion or hand off. - After fixing findings, rerun the required checks and review the changes again.
- If an automatic review finds
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

