Skills

Which of a CLI's skills an agent is loaded with — a setting of the agent, so two agents of one flow may be loaded differently, and neither touches the settings of the CLI itself.

ClaudeCodeAgentConfig(model=…, effort=…, skills=("code-review", "run"))

Value
None — the default	the CLI as it comes, which is every skill it finds
a tuple	exactly those and nothing else, whatever is installed afterwards

Why the tuple is exact

Every backend is told the other way round: a CLI comes with its skills loaded and has to be talked out of one. So what actually goes on the wire is the rest of them, worked out by looking at what is installed:

from hmz.agents.skills import leaving, skills

skills("claude")                       # what it would load here: yours, and this project's
leaving("claude", ("code-review",))    # what to switch off so that only that one is left

An agent that has been asked has exactly those skills from then on — including against a skill installed tomorrow.

What each backend can be told

Backend	How	What it comes to
claude	--disallowedTools "Skill(<name>)"	the agent is refused the skill. Claude still lists it — no flag takes one off that list
codex	-c skills.config=[{name="…", enabled=false}] on its app server	the skill is not loaded for that server; the user's own config.toml is untouched
kimi	—	kimi web takes no --skills-dir, so a skill it finds is one it loads
pi	—	it is told which skills to load by path, and finds none of its own to choose between
opencode, mimo	—	neither offers a way of switching one off for a single run

A backend with no way of being told anything is one where the setting has nothing to do. That is why the interface offers the choice only where it means something: a list to choose from that nothing acts on is a list that lies.

At the prompt

On the model step of /agents, ctrl+s. A side question about the same agent, so a key rather than a step of its own; the tuning line reads ◉ every skill · ctrl+s to choose.

   ❯ 1. [✔] code-review    Review the current diff… (yours)
     2. [ ] dataviz        Use this skill whenever you… (yours)
     3. [✔] housekeeping   Tidies the tree (this project)

• The skills are found where the CLI itself looks — yours and this project's — read for the name and the line each describes itself with. Nothing is asked of the CLI, which would mean starting it.
• Every box starts ticked, which is how a CLI comes.
• space switches the one under the cursor, enter takes the lot, esc leaves the agent loaded as it was.

Where each CLI keeps its skills is written down in hmz.backends:

Backend	Yours	This project's
claude	~/.claude/skills/*/SKILL.md	.claude/skills/*/SKILL.md
codex	~/.codex/skills/*/SKILL.md, ~/.agents/skills/*/SKILL.md	.agents/skills/*/SKILL.md, .codex/skills/*/SKILL.md

On a command line

No -a spells a skill list. An agent loaded with a particular set is one built in Python and handed to Runner, or one set up on the /agents sheet — where the choice is remembered per agent, per flow, per project.

from hmz.agents import ClaudeCodeAgent, ClaudeCodeAgentConfig
from hmz.runner import Runner

reading = ClaudeCodeAgentConfig(model="claude-opus-5", effort="high", skills=("code-review",))
writing = ClaudeCodeAgentConfig(model="claude-opus-5", effort="high")

Runner("official/rlar", [
    ClaudeCodeAgent(writing, name="actor"),
    ClaudeCodeAgent(reading, name="reviewer"),
]).run("$(cat TASK.md)")

The reviewer reading a change need not be carrying what the builder writing it was.

See also

• Permissions — the other per-agent narrowing
• Agents › Which skills an agent is loaded with
• TUI › What each agent is loaded with