Usage
03 / 05
scirix-cli is built to be driven by an AI agent. The installer ships
an agent skill that teaches Claude Code, Codex, GitHub Copilot,
Gemini CLI, Grok Build, Hermes Agent, JetBrains Junie, Kimi Code CLI,
Qwen Code, OpenCode, and OpenClaw how to authenticate, pick a
namespace, and query the Reality Model on your behalf. Direct shell use works fine too — the same
commands, with no agent in the loop.
Supported agents
Full — the agent skill plus a working pre-prompt hook that injects Reality Model context into every prompt.
Limited — the skill works, but per-prompt injection is constrained on the agent's side.
| Agent | Support | Notes |
|---|---|---|
| Claude Code | Full | Terminal, VS Code, and JetBrains — the extensions share ~/.claude |
| Codex | Full | Terminal, VS Code, and JetBrains — all surfaces share ~/.codex |
| Copilot (VS Code / Visual Studio) | Full | |
| Copilot CLI | Limited | Its prompt event is notification-only; the hook fires but can't inject |
| Copilot in JetBrains IDEs | Limited | Skills load in agent mode; Copilot's hooks don't run in the IDE surface |
| Gemini CLI | Full | Gemini Code Assist is a separate product and isn't covered |
| Qwen Code | Full | |
| Junie CLI (JetBrains) | Full | |
| Junie in JetBrains IDEs | Limited | Skill only — Junie's hooks run in its CLI, not the IDE plugin |
| Kimi Code CLI | Full | |
| OpenCode | Full | Via a generated plugin; the VS Code companion wraps the same process |
| OpenClaw | Full | Generated plugin; needs a one-time enable + usually --namespace |
| Hermes Agent | Full | Generated plugin; needs a one-time enable + usually --namespace |
| Grok Build | Limited | Skill only — its prompt event ignores hook output |
See Install for how the skill and hooks land in each agent, and the per-agent setup notes.
Driving the CLI from an agent
After install, you can ask your agent things like:
"What's in the auth flow for this codebase?"
"List dependents of
MyApp.Auth.LoginService.""Are there any open conflict findings right now?"
The agent reads the installed skill, runs the right scirix commands,
and synthesises the output. You don't have to remember subcommand
shapes — the skill carries that knowledge.
Where the skill lives
The installer mirrors the bundled skill to:
| OS | Path |
|---|---|
| Linux / macOS | ~/.scirix/skills/scirix/ |
| Windows | %USERPROFILE%\.scirix\skills\scirix\ |
It then symlinks (Linux/macOS) or junctions (Windows) that directory into each detected agent's skills location:
| Agent | Linked into |
|---|---|
| Claude Code | ~/.claude/skills/scirix/ |
| Codex | ~/.codex/skills/scirix/ |
| Copilot | ~/.copilot/skills/scirix/ |
| Gemini CLI | ~/.gemini/skills/scirix/ |
| Grok Build | ~/.grok/skills/scirix/ |
| Hermes | ~/.hermes/skills/scirix/ |
| Junie | ~/.junie/skills/scirix/ |
| Kimi Code | ~/.kimi-code/skills/scirix/ |
| Qwen Code | ~/.qwen/skills/scirix/ |
| OpenCode | ~/.config/opencode/skills/scirix/ |
| OpenClaw | ~/.openclaw/skills/scirix/ (copied, not linked) |
Because every agent points at the same source, upgrading scirix-cli
upgrades the skill in every agent at once — there's no per-agent state
to keep in sync.
Refreshing the link
If you install an agent after scirix, re-run the installer (or run the link step manually):
ln -sfn ~/.scirix/skills/scirix ~/.claude/skills/scirix
New-Item -ItemType Junction -Path "$env:USERPROFILE\.claude\skills\scirix" `
-Target "$env:USERPROFILE\.scirix\skills\scirix"
Driving the CLI directly
Everything the agent can do, you can do from the shell — useful for scripts, CI, and quick one-offs. The CLI is self-documenting:
scirix help
scirix rm --help
scirix auth --help
--help works on every subcommand, and JSON is the default output
format across scirix rm so it pipes straight into jq or downstream
tools.
Cloning every repo in a namespace
For the common "I'm new on this machine, give me every repo in the namespace" case:
mkdir ~/projects/acme && cd ~/projects/acme
scirix ns pull
That clones every repository registered in the active namespace into
the current directory. Future runs from anywhere under the same tree
do a git pull --ff-only on each repo instead. Auth (SSH keys,
credential helper) is whatever your system git is configured with.
Next
Configuration covers state on disk, env-var overrides, and pointing the CLI at a non-default API host.