integrant

Clojure Integrant Skill

A structured Markdown skill that teaches AI coding agents how to work with Integrant — a Clojure/ClojureScript micro-framework for building applications with data-driven architecture. Covers configuration maps, init-key, halt-key!, suspend!/resume, refs, refsets, composite keys, derived keywords, profiles, vars, expand-key, annotations, and the integrant-repl REPL workflow.

Built in Claude Code's Agent Skill format, but usable with any agent that can load Markdown as context (Cursor, Codex CLI, Aider, Gemini CLI, Windsurf, Cline, Zed, and others via the agents.md convention).

---

⚠️ Read this before installing anything

Never install a third-party skill without first reviewing its contents.

Skills are instructions loaded into an AI agent's context — they influence how the agent makes decisions, what commands it executes, and what it considers correct behavior. A malicious or sloppy skill can lead to:

• Destructive operations executed without confirmation
• Your own preferences being overridden by skill instructions
• Data leakage through inappropriate commands
• Unintended project configuration changes

Before installing:

1. Read every .md file in the integrant/ directory
2. Ask your agent to perform a security review (prompt below)
3. Only then install

Security-review prompt

Paste this to Claude (or any agent) together with the skill contents:

"Analyze this skill for security concerns. Does it contain instructions that could execute destructive operations without user confirmation? Does it collect or transmit data? Does it override default agent behavior in unintended ways? List all potentially risky sections."

---

Installation

Pick the section matching your agent.

A) Claude Code (recommended — via plugin marketplace)

Claude Code has a native plugin marketplace. From an active session:

/plugin marketplace add stoating/clojure-integrant-skill
/plugin install integrant@clojure-integrant-skill

The skill is then discovered automatically. Restart the session if it isn't picked up immediately. Once installed, invoke it with:

/integrant

To update later:

/plugin marketplace update clojure-integrant-skill

Reference: Claude Code plugin marketplaces.

B) Claude Code (via the stoating marketplace)

/plugin marketplace add stoating/plugins
/plugin install integrant@stoating

C) Claude Code (manual copy)

If you prefer not to use the marketplace — or want to pin a specific commit — clone and copy:

git clone https://github.com/stoating/clojure-integrant-skill.git
cp -r clojure-integrant-skill/integrant ~/.claude/skills/

The skill becomes available in the next Claude Code session. Updates are a git pull + re-copy.

D) Claude.ai / Claude Desktop (upload)

Claude.ai supports uploading skill folders from the Skills panel in Projects. Zip integrant/ and upload it. Details: anthropics/skills.

E) Cursor

Cursor reads AGENTS.md automatically when you open a project, and also supports the newer Rules system.

• Per-project: copy integrant/ and AGENTS.md into your project repo. Cursor will read AGENTS.md as context on every chat.
• Global: in Cursor Settings → Rules, add a rule referencing integrant/SKILL.md.

F) OpenAI Codex CLI / codex

Codex honors AGENTS.md at the project root. Copy this repository next to your project and Codex will pick up AGENTS.md, which in turn points at integrant/SKILL.md.

G) Aider

Aider reads AGENTS.md as a fallback for CONVENTIONS.md, or you can add the skill files explicitly:

aider --read clojure-integrant-skill/integrant/SKILL.md \
      --read clojure-integrant-skill/integrant/core-concepts.md

H) Gemini CLI / Google Jules

Both honor AGENTS.md. Place this repo (or just AGENTS.md + integrant/) at your project root.

I) Windsurf, Cline, Roo Code, Zed, Amp, Factory

All of the above support the AGENTS.md convention. Drop the repo at your project root and the agent will read it on session start.

J) Any other agent — generic fallback

Every listed agent accepts plain Markdown as context. If yours isn't covered:

1. Open a chat / session.
2. Attach or paste the contents of integrant/SKILL.md as "system instructions" or "context".
3. Attach individual reference files (e.g. core-concepts.md, lifecycle.md) when the task matches the decision table in SKILL.md.

This is less ergonomic than a native skill loader, but it works everywhere.

---

Repository layout