Commands inside your agent
hero setup installs 12 Ekamyth Agent OS commands into your agents: slash commands in Claude Code, and skills of the same names in Codex.
What the Agent OS commands inside your agent are
Each one lets you ask your agent about Agent OS without leaving the session. In Claude Code you type it as a slash command, such as /hero:health. In Codex each is a skill named hero-<name>, such as hero-health. Each matches a hero command you can also run in a terminal.
The 12 commands
| Claude Code | Codex | What it does | Terminal command |
|---|---|---|---|
/hero:health | hero-health | One read of how your agent is doing across rules, cost, proof and records, with the top risks and what to run next. | hero health |
/hero:enforce | hero-enforce | Every rule Agent OS knows, and whether each agent enforces it. | hero enforce |
/hero:codify | hero-codify | The older name for /hero:enforce. It does the same thing. | hero enforce |
/hero:govern | hero-govern | Agent OS's rules for risky actions, such as deleting, deploying, paying or sending, including the ones that wait for your approval. | hero govern |
/hero:attest | hero-attest | The record of what your agent did, kept on your machine as a hash chain, without the content of prompts or replies. | hero attest |
/hero:trace | hero-trace | What your agent did, step by step, with a plain verdict and the reason for it. | hero trace |
/hero:efficiency | hero-efficiency | Why your agent's work costs what it does, and which fix would lower it. | hero efficiency |
/hero:route | hero-route | Which model calls moved to another model and why, where the tokens went, and which agents reach the model gateway. | hero route |
/hero:memory | hero-memory | How your rules and context load into a session within a budget, and what was left out. | hero memory |
/hero:eval | hero-eval | Scores your agent against a baseline, so you can check before you change a model, prompt or tool. | hero eval |
/hero:brain | hero-brain | Saves, lists and resumes local work checkpoints across Claude Code and Codex. | hero brain threads |
/hero:identity | hero-identity | Shows or enrolls the one local agent identity that Claude Code and Codex share. | hero identity |
Where hero setup installs them
| Agent | Location |
|---|---|
| Claude Code | ~/.claude/commands/hero/<name>.md |
| Codex | ~/.agents/skills/hero-<name>/SKILL.md, and the same file under Codex's own skills folder, ~/.codex/skills/hero-<name>/SKILL.md by default |
The Codex copies are installed only when Codex is on the machine. Setup records each file it installs, so it can check and remove them later.
Install only these commands
This command will:
- Install or update the 12 commands in Claude Code and, if it's installed, Codex. It changes no other agent setting.
hero setup commands
Output from version 0.1.36rc13 (build cd08bd39c026), on Linux, with paths shortened to ~.
hero setup: installed 12 slash commands → ~/.claude/commands/hero
/hero:health
/hero:codify
/hero:govern
/hero:efficiency
/hero:route
/hero:memory
/hero:attest
/hero:trace
/hero:eval
/hero:brain
/hero:identity
/hero:enforce
hero setup: installed 12 Codex skills → ~/.agents/skills and ~/.codex/skills
digests recorded → ~/.hero/skills-install.json
Restart Claude Code and Codex (or open a new session) to pick up the installed adapters.
Check the commands are in place
hero setup --check
It reports a command file that's missing or no longer matches what setup installed. Older duplicates in ~/.agents/skills/ are listed as stale duplicate (not removed); setup never deletes them.
Output from version 0.1.36rc13 (build cd08bd39c026), on Linux, when the commands are in place. Trimmed where you see …. The address is the one setup pointed your agents at, so yours can differ.
…
hero setup --check — proxy http://127.0.0.1:50217
…
✓ claude covered — env.ANTHROPIC_BASE_URL targets Ekamyth Runtime · 5/5 hooks wired · …
…
✓ codex covered — model_providers.hero_runtime.base_url targets Ekamyth Runtime · 6/6 hooks wired · …
…
session memory: wired (claude, codex)
all detected agents route through Ekamyth Runtime ✓
skill adapters match the packaged digests ✓
The last line is the one about the commands. The run did not show a missing, changed or stale command file, so this page has no output for those.
Remove the commands
Run this in your own terminal:
hero setup --undo
It removes each command file whose content still matches what setup installed. A file you've edited is yours, and undo leaves it in place. Undo also takes the rest of Agent OS out of your agents; see hero setup.
Limits of the commands inside your agent
| Limit | Behaviour | Can you change it? | What to do |
|---|---|---|---|
| Agents | Claude Code and Codex only. | No | Run the hero command in a terminal. |
| Open sessions | A session that was already open may not see newly installed commands. | No | Restart your agent, or open a new session, to pick up the commands. |
| Edited files | Undo leaves a command file you've edited. | Yes | Delete it yourself. |
Related
Checked against Agent OS 0.1.36rc14 (build 599c5e120c18) on 17 Nov 2026.