A skill is a set of instructions your agent follows. Magistr installs nine project skills for supported Codex and Claude Code setups. Connect the agent and open a fresh session before using them.
Enter $name in Codex or /name in Claude Code. These are conversation
commands, not shell commands. A portable MCP connection exposes tools but does
not promise the same skill invocation syntax in every host.
Available skills
| Name | Use it to | Starts implementation? |
|---|---|---|
init |
Interview you and tailor project setup to the stack and workflow. | No |
craft |
Apply the project’s usual planning, evidence, delivery, and verification practices to your request. | Only within the work you requested |
next |
Resume interrupted roadmap work or finish one ready chunk. | Yes |
roadmap |
Implement one ready roadmap chunk. | Yes |
next-loop |
Finish one chunk, save a bounded-run handoff, then stop for a fresh session. | Yes, one chunk per invocation |
roadmap-refresh |
Research and revise a focused part of the roadmap. | No |
signals |
Investigate feedback and surface or promote supported ideas. | No |
validate |
Run the configured standard and required custom quality gates. | No |
workspace-routing |
Use Magistr’s source search, reading, editing, commands, and undo tools. | Supports the requested work |
Make the scope explicit
For planning without implementation, ask for roadmap-refresh and name the
area. For one implementation, use next. The roadmap skill is also an
implementation workflow despite its name.
For example, in a Codex conversation:
$roadmap-refresh
Review the import workstream against the current code. Update its plan;
do not implement features.
Replace “import” with an actual workstream in your project. The expected result is recorded research and an updated plan.
Understand the loop
The chat next-loop skill completes one chunk and prepares a handoff. A new
session must read that handoff and saved state before the next invocation.
It does not launch an unlimited sequence inside the same conversation.
The terminal command magistr next-loop --max-iterations 3 instead manages
up to three fresh agent sessions. See Working in chunks.
Find missing or outdated skills
Codex project skills live in .agents/skills/; Claude Code project skills
live in .claude/skills/. Use ./.codex/run for Codex’s project selection.
An already-open host may need a fresh session after setup changes.
In the project terminal:
magistr doctor .
magistr repair . --dry-run
Review the repair report before applying it with magistr repair ..
Customizations may be preserved instead of overwritten; inspect preserved files
if the behavior differs from the current guide.
Skills guide the agent. Required checks and permissions still apply, and a skill invocation does not establish that all of its steps succeeded.