Cross-Tool Skill Library: alirezarezvani/claude-skills on 13 Hosts

A single collection of 362 production-ready Claude Code skills, agents, and plugins that ships with conversion scripts for 13 different AI coding tools. alirezarezvani/claude-skills targets the cross-tool gap: one curated bundle, installable on Claude Code, Codex, Cursor, Windsurf, Aider, and others, covering engineering, marketing, product, compliance, C-level advisory, research, and project management.

Why This Skill Matters

Most skill collections target one host. The moment you switch tools — from Claude Code to Cursor, or from Codex to Windsurf — you re-install, re-curate, re-learn. This collection ships conversion scripts (./scripts/convert.sh --tool all) so the same 362 skills render correctly on each host's expected format. That saves the duplication and the version drift between copies.

The breadth is the other story. Beyond engineering (52 core + 84 "POWERFUL Tier" skills), the collection covers domains where most skill lists stop: marketing (48), product (17), C-level advisory (68), regulatory and quality management (19), compliance (9), finance (4), and research operations (14). You pull in a startup-cto persona for architecture decisions, a growth-marketer for a launch plan, a compliance skill for a SOC 2 walkthrough — without leaving your agent host.

Installation

Installation differs by host. Pick yours:

For Claude Code:

/plugin marketplace add alirezarezvani/claude-skills
/plugin install engineering-skills@claude-code-skills

For Codex:

npx agent-skills-cli add alirezarezvani/claude-skills --agent codex

For Cursor, Windsurf, Aider, and other tools:

./scripts/convert.sh --tool all
./scripts/install.sh --tool <name> --target .

Replace <name> with the tool you target (cursor, windsurf, aider, ...). Some shipped tools — for example landing_page_scaffolder — require Node.js downstream even though the CLI scripts use stdlib Python 3 only. Confirm your environment before relying on those.

Real Workflow: Audit a Third-Party Skill Before Installing

A realistic scenario, not "hello world": you found a promising skill on a marketplace and want to know whether it's safe to install. Use the bundled skill-security-auditor:

python3 engineering/skill-security-auditor/scripts/skill_security_auditor.py /path/to/skill/

Expected output: a structured report flagging any prompt-injection patterns, unsafe shell calls, network exfiltration, missing provenance metadata, and skill-bundle permissions that exceed what's typical. Treat the report as a triage signal, not a guarantee — the auditor reads what the skill contains, not what the upstream publisher's track record is.

Pair the auditor with a pin: install the skill at a specific version, never @latest. The collection's own README notes that semantic-versioning is honored for patches, but cross-cutting improvements can still require reconfiguring tool-specific rules in Cursor/Windsurf after upgrading — pin so you can reproduce.

Real Workflow: Combine Personas and Skills for a Product Launch

For a multi-domain task, the collection's "persona + skill" pairing is what makes it work.

A realistic example — a 6-week product launch plan:

1. Activate the startup-cto persona for technical scope.

2. Activate growth-marketer for launch-channel planning.

3. Combine with skills from engineering/ (e.g. engineering-fullstack-architect), marketing/ (e.g. marketing-content-strategist), and analytics/ for measurement.

4. Prompt your agent with the launch goal, the personas, and the skills:

   Use the startup-cto and growth-marketer personas with
   engineering-fullstack-architect, marketing-content-strategist,
   and the analytics skill. Plan a 6-week product launch for
   a developer-tools product targeting senior backend engineers.
   

5. Walk the output. Because each skill encodes its domain's conventions, the agent reasons over the right shape (technical scope, content calendar, measurement plan) rather than guessing.

Expected output: a plan that interleaves technical milestones with launch beats and measurement checkpoints — not a generic Gantt chart dressed up.

Tips

• Pick by tool, not by skill count. The README notes the visible count varies (330 vs 362) depending on which install path resolves. Don't compare counts across tools; compare what's installed in your environment.
• Install by domain, not by everything-at-once. 362 skills is too much for one context window. Activate engineering first if that's your work; add marketing when you need a launch.
• Run the security auditor on new skills, even from this collection. Marketplace supply-chain risk is real — the auditor catches content, not provenance.
• Pin versions. Cross-cutting improvements can land without a major bump; pinning prevents surprise reconfiguration after upgrade.

When Not to Use This

• You need certified advice. The compliance, C-level advisory, and finance skills are scaffolds for thinking — not certified legal, financial, or regulatory advice. Always have a qualified human review the output.
• You only use one tool, one domain. If you're a single-host user working in engineering, a smaller targeted collection will give you less context noise.
• You can't pin versions. Without version pinning, marketplace upgrades can silently change which skills are resolved. If your environment doesn't support pinning, this collection's frequent updates become a liability.
• You need offline / air-gapped operation. Several skills call external APIs (analytics, marketing tools). The 644 stdlib-Python tools run offline, but the persona + skill combinations typically don't.

See the leaderboard for more skills.