CDLI Agent Skills

CDLI Agent Skills is a small public collection of capability-first engineering skills by CDLI. It distills internal agent prompt patterns into portable, readable workflows for auditing, mapping, refactoring, and shipping software with evidence.

This is a visibility and craft collection, not a paid product. Each skill is intentionally narrow: it says what task should trigger it, how to perform that task, and what output shape to produce.

Included Skills

• reviewing-code: review pull requests and diffs for merge risk, correctness, performance, and maintainability.
• auditing-security: audit exposed surfaces for vulnerabilities, auth flaws, secret handling, and unsafe data flow.
• auditing-tech-debt: find high-interest debt, duplication, drift, dead code, and inconsistent implementation patterns.
• refactoring-code: restructure code for readability and maintainability without changing behavior.
• mapping-architecture: diagram and explain a codebase or system before making architecture judgments.
• reviewing-api-contracts: review REST and GraphQL contracts for resource shape, auth, errors, versioning, and developer experience.

Install

Clone the public repository, then copy the skill directories into the skills folder used by your agent runtime. Keep each SKILL.md with its parent directory name.

git clone https://github.com/cdliai/cdli-agent-skills.git

Claude Code personal skills:

mkdir -p ~/.claude/skills
cp -R cdli-agent-skills/*/ ~/.claude/skills/

Codex and other local agents:

mkdir -p ~/.codex/skills
cp -R cdli-agent-skills/*/ ~/.codex/skills/

Folder shape:

skills/
  reviewing-code/
    SKILL.md
  auditing-security/
    SKILL.md

Design Principles

• Convert personas into capabilities. A skill should trigger on a task, not an identity.
• Prefer evidence over assumption. Cite files, symbols, endpoints, or commands; mark anything unconfirmed.
• Keep reports consistent. A predictable output format makes findings easier to compare, share, and act on.
• Keep public skills portable. Do not include private endpoints, internal paths, secrets, or surprise side effects.

Part of CDLI Agent Skills by CDLI - https://cdli.ai