Claude Code Skills

Reusable skills for Claude. Use them as slash commands in Claude Code or as project instructions in claude.ai.

Skill	What it does
commit	Atomic git commits with conventional messages. Stages specific files, writes a subject + prose body explaining non-obvious decisions. Never uses git add -A.
code-review	Adversarial multi-persona code review. A panel of reviewers finds 3-10 real issues across code quality, architecture, risk, and user impact. No compliments.
research	Multi-round deep research with source verification. Iterates until the question is answered, cross-verifies claims, flags single-source and stale information.
grill-me	Relentless questioning on any topic until all decision branches are resolved. Plans, designs, code, strategy — anything.
mock-interview	Simulates a real job interview with 4 rounds: behavioral, technical, code walkthrough, live problem solving. Adapts to any role.
persona-bai-ivan	Bulgarian grumpy mentor persona. Completes every task but delivers feedback with attitude and slang.

Setup

Claude Code (CLI)

Copy the skills into your Claude Code skills directory:

# Clone this repo
git clone https://github.com/aslavchev/claude-code-skills.git

# Copy skills to your Claude Code config
cp -r claude-code-skills/skills/* ~/.claude/skills/

Restart Claude Code. The skills appear as slash commands.

claude.ai (web client)

1. Go to claude.ai/projects or click Projects in the left sidebar
2. Click + New Project and give it a name
3. Click Set project instructions
4. Copy the content of any SKILL.md file (skip the YAML frontmatter between the --- lines) and paste it as the instructions
5. Click Save instructions
6. Start a chat within the project and tell Claude what you need:
   • "Review this code" and paste your code
   • "Research Playwright vs Cypress for API testing"

The /commit skill requires terminal access and only works in Claude Code.

Usage

Claude Code

Each skill is invoked with /skill-name followed by optional arguments. Arguments are whatever you type after the command — file paths, questions, flags. If you invoke a skill without arguments, it will ask you what it needs before proceeding.

/commit src/utils/format.ts
/commit                              → asks which files to include

/code-review src/pages/cart.tsx
/code-review --staged                → reviews only staged changes
/code-review                         → asks which files to review

/research how does GitHub Actions caching work
/research --save                     → asks the question, saves report to file
/research                            → asks what to research

claude.ai

Use natural language. The skills work as instructions that shape how Claude responds:

"Review this code: [paste code]"
"Research how companies are adopting AI in CI/CD pipelines"

Skill details

/commit

Enforces atomic, conventional commit discipline:

• One logical change per commit
• type(scope): subject format with imperative mood
• Prose body (no bullets) only when the "why" isn't obvious
• Shows the commit for approval before executing
• Never skips pre-commit hooks

/code-review

An adversarial review panel, each reviewer with a different lens:

• Linus Torvalds — simplicity, naming
• Staff Engineer — architecture, patterns
• Principal Anthropic Developer — idiomatic design
• Principal FAANG Engineer — production readiness
• Munger & Marks — inversion, risk
• Jeff Bezos — user impact
• Claude Shannon — signal vs noise, redundancy

Auto-detects language and framework. Works on any codebase.

/research

Iterative web research that goes deeper than a single search:

• Searches from multiple angles, reads full sources
• Cross-verifies claims across independent sources
• Flags single-source claims, contradictions, and stale data
• Produces a cited report with open questions
• Depth scales automatically with question complexity

/grill-me

Relentless questioning on any subject:

• One question at a time, waits for your answer
• Follows up on vague or incomplete answers
• Explores the codebase when questions can be answered by reading code
• Summarizes resolved decisions at the end

/mock-interview

Four-round interview simulation:

• Round 1 — Behavioral (STAR format, probes for specifics)
• Round 2 — Technical knowledge (adapts to your stack)
• Round 3 — Code walkthrough (challenges your decisions)
• Round 4 — Live problem solving (watches you think)
• Ends with hire/no-hire verdict and specific improvement areas

/persona-bai-ivan

Bulgarian mentor persona that reviews code with attitude. Speaks your language for code, Bulgarian for personality. Uses slang from a curated reference table.

License

MIT