By hackberrybay
Hackberry's AI tooling toolkit. General-use Claude Code skills, agents, hooks, and rules — plus a place for Cursor and shared resources. Currently includes the code-quality + security review suite and a self-improving rule system.
General TypeScript/JavaScript code quality reviewer. Reviews for type safety, logic correctness, error handling, function design, and best practices. Use for any .ts/.tsx/.js/.jsx file.
OWASP-focused security reviewer. Checks for injection, auth gaps, sensitive data exposure, input validation, and misconfigurations. Use for security-critical files or before major releases. Uses Opus for thorough analysis.
Self-improvement agent. Extracts recurring issue patterns from review output and manages the recurring-issues log. Called by /review-pr to log patterns, and by /improve-rules to promote them into permanent rule files.
Reads the recurring-issues log and promotes issues that have appeared 3+ times into the permanent rule files (typescript.md or security.md). Run periodically after multiple review cycles — typically once per sprint.
Generates a pull request title and description from the current branch diff. Reads commit messages and changed files to produce a consistent, structured PR description ready to paste into GitHub.
Reviews a single TypeScript/JavaScript file for code quality, best practices, and rule violations. Pass the file path as the argument.
Reviews all files changed in the current branch vs a base branch. Auto-routes each file to the right reviewer, runs PR-level checks, logs recurring patterns, and outputs a verdict. Run before pushing or opening a PR.
OWASP-focused security review of a file or directory. Checks for injection, auth gaps, sensitive data exposure, input validation failures, and misconfigurations. Uses Opus for thorough analysis. Run before major releases or on security-critical code.
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Hackberry's shared AI tooling. General-use Claude Code skills, agents, hooks, and rules — plus space for Cursor rules and tool-agnostic prompts. One install per developer, available in every project.
Built for the Hackberry stack (NestJS, React Native/Expo) but applicable to any TypeScript project.
/review-pr, /review-file/review-security (Opus-backed, OWASP-focused)/pr-description/improve-rules promotes recurring issues into permanent rule filesFuture additions live alongside: Cursor rules under cursor/, tool-agnostic prompts under shared/.
Claude Code clones plugins via SSH. If you have SSH keys configured for GitHub, no extra setup is needed.
If you don't have SSH keys set up and get a Permission denied (publickey) error during install, run this once to fall back to HTTPS:
git config --global url."https://github.com/".insteadOf "[email protected]:"
Note: this redirects all GitHub git operations on your machine to HTTPS. Skip it if you use SSH keys.
Then install:
/plugin marketplace add https://raw.githubusercontent.com/hackberrybay/hb-ai-toolkit/main/.claude-plugin/marketplace.json
/plugin install hb-ai-toolkit@hb-marketplace
After this, all skills (/review-pr, /review-file, etc.) are available in every project without any --plugin-dir flag.
To enable auto-updates (recommended), do this once after installing:
/plugin
Go to Marketplaces → select hb-marketplace → enable auto-update. Claude Code will then update the plugin automatically on startup.
To update manually:
/plugin update hb-ai-toolkit
Bump the version in .claude-plugin/plugin.json with every meaningful change so it's clear an update is available.
If you want to edit and iterate locally without going through the plugin system:
git clone [email protected]:hackberrybay/hb-ai-toolkit.git
cd hb-ai-toolkit
./scripts/install.sh claude
Skills/agents/rules are symlinked into ~/.claude/. claude/settings/ snippets are not auto-merged — copy what you want into ~/.claude/settings.json.
./scripts/install.sh cursor
shared/ is reference material — copy what you need into a project, or read it directly.
| Command | Model | What it does |
|---|---|---|
/review-pr [base] | Sonnet | Reviews all changed files vs base branch (default: main). Gives verdict: APPROVE / REQUEST CHANGES / NEEDS DISCUSSION |
/review-file <file> | Sonnet | Reviews a single file for quality and best practices |
/review-security [path] | Opus | OWASP-focused security review of a file or directory |
/improve-rules | Sonnet | Promotes recurring issues (3+ occurrences) into permanent rule files |
/pr-description | Haiku | Generates a PR title and description from the current branch diff |
The plugin uses different models for different tasks to balance quality and cost:
| Task | Model | Why |
|---|---|---|
| Code quality review | Sonnet | Good quality at reasonable cost |
| Security review | Opus | High-stakes — worth the cost for thorough analysis |
| PR descriptions, rule promotion | Haiku / Sonnet | Low complexity, Haiku is sufficient |
To override for a session:
claude --model opus # force Opus everywhere for this session
The review skills get smarter over time without manual rule writing.
/review-pr reviews your branch and, at the end, checks if any issue class appeared 3+ times in this run.
If so, it appends the pattern to claude/rules/recurring-issues.md.
Every sprint or two, a team member runs /improve-rules.
Any issue with 3+ total occurrences across sessions gets promoted into the permanent rule files (claude/rules/typescript.md or claude/rules/security.md).
On the next review, the promoted rule is applied automatically.
claude/rules/typescript.md and claude/rules/security.md are plain markdown — edit them directly and commit.
Rule changes go through normal PR review, so the team stays aligned on what standards are actually enforced.
claude/rules/recurring-issues.md is auto-managed. Do not edit it manually.
This toolkit uses a team marketplace — a marketplace.json hosted in this repo's main branch. It is the source of truth for the team install.
npx claudepluginhub hackberrybay/hb-ai-toolkit --plugin hb-ai-toolkitComprehensive code review with parallel specialized agents covering security, architecture, performance, and more
使用多個專門代理進行自動化程式碼審查,配備基於置信度的評分系統以過濾誤報
Comprehensive code review with architecture, security, and PR enhancement
Code quality enforcement, architecture skills, structural code graph, and impact-aware reviews.
Automated code review, security scanning, and quality enforcement
Reviews pull request changes to provide feedback, check for issues, and suggest improvements before merging into the main codebase.