Guides

Use AI skills

Install Dibbla skills into Claude Code, Cursor, Gemini CLI, and other agent-aware editors.

A skill is a bundle of Markdown files that teaches an AI coding agent how to use a tool — what commands exist, when to call them, what edge cases to watch for. The Dibbla skills ship inside the dibbla binary, so dibbla skills install does not require network access and gives you the version matching your CLI. They are additionally published over HTTP, for agents that have no CLI — see Fetching the skill without the CLI.

If you ran dibbla init, the dibbla skill is already installed in the current project. The rest of this page covers how it’s used by each agent, when to install at $HOME scope instead, what’s bundled, and how to clean up.

Install it

Inside the project directory:

$dibbla skills install dibbla
writing .claude/skills/dibbla/SKILL.md
writing .claude/skills/dibbla/examples.md
writing .claude/skills/dibbla/guardrails.md
writing .claude/skills/dibbla/reference.md
updated AGENTS.md
updated GEMINI.md
✓ installed skill "dibbla" into the current project

# Install to $HOME — applies to every project on this machine
$dibbla skills install dibbla --user

Pick your agent to see how to verify it works:

Code agent

Claude Code reads .claude/skills/<id>/SKILL.md natively. Open the project in Claude Code and try:

Read the dibbla skill, then tell me what this project still needs before it can be deployed to Dibbla.

Claude will read the skill and answer from it — naming the handbook and the pre-deploy review the gate requires, rather than guessing. That is the signal the skill was picked up.

This deliberately asks a question instead of deploying: a verification step should not ship anything, and dibbla deploy would be refused here anyway until the handbook and REVIEW.md exist.

Cursor reads AGENTS.md at the project root. Start a new chat and try:

Use the dibbla skill to tell me what this project still needs before it can be deployed to Dibbla.

If Cursor doesn’t pick it up, reload the workspace — AGENTS.md is indexed on session start.

Gemini CLI reads GEMINI.md at the project root. Run gemini inside the project and try:

Use the dibbla skill to tell me what this project still needs before it can be deployed to Dibbla.

Opencode reads AGENTS.md at the project root. Start a fresh session and try:

Use the dibbla skill to tell me what this project still needs before it can be deployed to Dibbla.

Most modern agents — Codex, Copilot, Windsurf, Aider, Zed, Warp, RooCode, others — follow the AGENTS.md open standard. The pointer block dibbla skills install writes is enough for any of them to find and use the skill.

If your agent isn’t AGENTS.md-aware yet, pipe the skill into its context manually:

dibbla --skill-prompt

That command prints the same instructions written for direct LLM consumption — paste the output into your agent’s system prompt.

Per-project vs machine-wide

Scope	Where it lives	When to use
Per-project (default)	`.claude/skills/`, `AGENTS.md`, `GEMINI.md` at the project root	The repo is shared with teammates. Commit the skill files so everyone gets the same agent behavior — no per-person setup.
Machine-wide (`--user`)	`$HOME/.claude/skills/`, `$HOME/AGENTS.md`, `$HOME/GEMINI.md`	You work across many repos and don’t want to repeat the install. Skill applies to every project unless a per-project install overrides it.

The two modes don’t conflict — per-project files take precedence when an agent finds both.

Bundled skills

Skills are versioned with the CLI. Upgrade dibbla to get newer versions. The HTTP mirror is pinned the same way — it serves a tagged release, never a floating latest.

$dibbla skills list
ID                  DESCRIPTION
dibbla              Use the Dibbla CLI to scaffold, deploy, and manage apps
dibbla-ai-gateway   Point local AI tools at the Dibbla AI gateway

dibbla — the main skill, installed by dibbla init. Covers deploys, apps, databases, secrets, manifests, workflows, and the Go SDK.
dibbla-ai-gateway — opt-in. Configures local AI tools (Claude Code, Opencode, Cursor, Cline, Windsurf, Zed) to route LLM calls through the Dibbla AI gateway so every call is captured under your org. Install separately: dibbla skills install dibbla-ai-gateway.

Idempotency

Re-running dibbla skills install is safe:

Files identical to the bundled version are no-ops — no mtime change, no rewrite.
Locally edited skill files cause the install to refuse to overwrite them. Pass --force to replace them, or keep your customizations in a separate file the installer doesn’t touch.
AGENTS.md and GEMINI.md use marker-delimited regions, so the installer can update its block in place without disturbing content you’ve added yourself.

Uninstall

dibbla uninstall --skill-only removes the skill files and strips the marker block from AGENTS.md / GEMINI.md at every recorded install root, leaving the rest of those files intact. Run it without --skill-only for a full cleanup (binary, keychain credentials, config directory, every install root).

Troubleshooting

Agent doesn’t see the skill after install. Restart the agent session — most editors index .claude/skills/, AGENTS.md, and GEMINI.md on startup.
AGENTS.md had unexpected content removed. The installer only edits the marker-delimited block. If you removed the markers manually, re-running dibbla skills install appends a fresh block instead.
Skill files were edited and you want the original back. Run dibbla skills install <id> --force to replace local edits with the bundled version.
You’re using an agent that doesn’t read AGENTS.md / GEMINI.md / .claude/skills. Use the bootstrap fallback: dibbla --skill-prompt prints the skill content for direct paste into a system prompt.
The agent has no dibbla binary at all. The same files are published over HTTP at https://dibbla.com/.well-known/agent-skills/index.json, with a sha256: digest per entry. See Fetching the skill without the CLI.

See the dibbla skills reference for the full flag surface, and the dibbla-skill.yaml schema for what each file contains.

← Previous

Logs

Use templates