dibbla skills

Install bundled skills that teach AI coding agents — Claude Code, Cursor, Gemini CLI, Opencode, Codex, and other AGENTS.md-compatible tools — how to drive the Dibbla CLI on your behalf.

A skill is a bundle of Markdown files that tells an AI agent how to use a tool — what commands exist, when to call them, what edge cases to watch for. The Dibbla skill ships inside the dibbla binary, so dibbla skills install does not require network access. The same files are also published over HTTP for agents that have no CLI — see Fetching the skill without the CLI.

dibbla skills install writes those files to your project (or $HOME) where Claude Code’s native skills system, plus the cross-tool AGENTS.md standard, and Gemini CLI’s GEMINI.md will pick them up automatically.

Supported agents

dibbla skills list

Print the skills bundled with this version of dibbla. Skills are version-locked to the binary — upgrade dibbla to upgrade your skills. The HTTP mirror is pinned the same way: it serves a tagged release, not a floating latest.

$dibbla skills list
ID       DESCRIPTION
dibbla   Use the Dibbla CLI to scaffold, deploy, and manage Dibbla apps

dibbla skills install

Write the skill files into the current directory (or $HOME with --user).

What gets written

Per-project install (default):

.claude/skills/dibbla/
├── SKILL.md          ← entry point
├── guardrails.md     ← the pre-deploy checklist
├── user-docs.md      ← writing the handbook a deploy requires
├── examples.md
├── reference.md
├── manifest.md
├── platform.md
├── workflows.md
├── sdk-go.md
└── ai-gateway.md
AGENTS.md     ← marker-delimited block; merges with existing content
GEMINI.md     ← same content for Gemini CLI

Ten files, not four — guardrails.md and user-docs.md in particular are what the CLI’s own deploy-gate error message points an agent at.

Both AGENTS.md and GEMINI.md use marker-delimited regions so re-running the install merges cleanly with content you’ve added yourself.

Examples

# Install into the current project
$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" (dibbla 1.2.33) into /Users/jane/projects/my-app

# Install to $HOME — applies everywhere
$dibbla skills install dibbla --user

$dibbla skills install dibbla --no-agents

# Re-apply the bundled skill, replacing local edits
$dibbla skills install dibbla --force

Idempotency

Re-running the install is safe:

• Files identical to the bundled version are no-ops — no mtime change, no rewrite.
• Files that have been edited locally cause the install to refuse to overwrite them. Pass --force to replace them, or add your own content in a separate file.
• Each file is written via temp + rename, so a killed process leaves no half-written state.
• CRLF vs LF line endings are preserved on round-trips.

Fetching the skill without the CLI

If an agent has no dibbla binary — a browsing agent, a CI job, someone evaluating Dibbla before installing anything — the same skill files are published over HTTP, following the Agent Skills Discovery RFC (Draft 0.2.0).

Start at the index:

curl -s https://dibbla.com/.well-known/agent-skills/index.json

{
  "$schema": "https://schemas.agentskills.io/discovery/0.2.0/schema.json",
  "skills": [
    {
      "name": "dibbla",
      "type": "archive",
      "url": "https://dibbla.com/.well-known/agent-skills/dibbla.tar.gz",
      "digest": "sha256:…"
    },
    {
      "name": "dibbla-faq",
      "type": "skill-md",
      "url": "https://dibbla.com/.well-known/agent-skills/dibbla-faq/SKILL.md",
      "digest": "sha256:…"
    }
  ],
  "_source": { "ref": "v1.2.49", "commit": "…" }
}

Verify the digest. It is in the index for a reason — it is what tells you the bytes a CDN handed you are the bytes we published:

curl -sL https://dibbla.com/.well-known/agent-skills/dibbla.tar.gz -o dibbla.tar.gz
sha256sum dibbla.tar.gz     # must equal the entry's `digest`, minus the "sha256:" prefix
tar -xzf dibbla.tar.gz      # → dibbla/SKILL.md + 9 reference files

If you only want the entry point, it is served uncompressed as well:

curl -s https://dibbla.com/.well-known/agent-skills/dibbla/SKILL.md

Which version am I getting?

The published files are mirrored from a tagged CLI release, never from main, and the index names that tag in _source.ref. So the HTTP path is version-pinned exactly like the binary is — _source.ref: "v1.2.49" means these are byte-for-byte the files that dibbla skills install dibbla writes from dibbla v1.2.49.

See also

• Available agents and the AGENTS.md standard
• dibbla run — what an agent typically calls after the skill is installed.