Skill chains

Link skills to run one after another — plan, then build, then review — and your agent runs the whole sequence as one skill.

What a chain is

A chain is an ordered list of skills you draw in Chains in the sidebar. It appears in Claude Code, Codex, Grok and AGY as one skill named chain-<name>. When your agent uses it, it gets each step's skill in turn, with your note for that step, and is told to finish each step before the next. Every step is still a signed skill, checked on your machine exactly as when it is used on its own.

Build a chain

1. Open Chains and press New chain.
2. Give it a name (lowercase, with dashes, like feature-flow), a title, and a line on what it's for. Your agent sees that line in its skill list.
3. Drag skills from the list onto the canvas, or press Add to put a skill after the last step and link it automatically.
4. Link steps by dragging from a step's right edge to the next step's left edge. To remove a link, select it and press Backspace.
5. Check the order shown under the canvas, then press Save chain.

The chain reaches your machines at the next skilvault sync, which runs by itself when a CLI session starts.

Using a chain in your CLI

Ask for it by name (“run the feature-flow chain”), or let your agent pick it from its skill list. You can also print it yourself:

$ skilvault read chain-feature-flow
# Skill chain: Feature flow

Run these 4 skills in order. …

## Step 1 of 4: brainstorming
…

skilvault list shows your chains after your pinned skills.

Your agent also finds chains on its own. skilvault find and the prompt hook (Claude Code and Codex) match your requests against your chains as well as the catalog, on your machine, and name a chain that fits, marked as your skill chain. Write What it's for the way a skill's description reads, saying when to use it (“Use when starting a new feature…”): that text is what it is matched on, together with the requests that led you to run it before.

Rules and limits

• A chain is one straight line: each step leads to at most one next step, and nothing loops back. The editor refuses a link that would fork, merge or loop.
• From 2 to 12 steps, and up to 50 chains per account.
• Every step must be a published skill your plan can read.
• Chains count towards Codex's and AGY's skill-list budget, and are listed before other skills.

When a step is unavailable

If a skill in a chain is unpublished later, or your plan changes, that step is named as unavailable when the chain is read, the other steps still run, and the command reports that the chain was incomplete. If your plan no longer includes chains, they leave your CLIs at the next sync; they are kept, and come back if you upgrade again.