DevStation CLI

Sessions & Commands

Talk to it in a session, give it one job from a script, or point it at a GitHub repository. Every run is saved as it goes, so nothing is lost if it stops.

A session

bash

cd your-project
devstation

devstation clears the terminal and opens its own session. A rule above each prompt shows the model, the folder and any mode that is on. Say what you want; the conversation carries across turns, so the second thing you ask can be about the first.

Key	Does
/ then Tab	Complete a slash command.
Ctrl-C	Interrupt the turn in progress.
Ctrl-D or /exit	Leave the session.

One job and out

bash

devstation run "fix the failing test in src/total.test.js, change the source not the test"

Good for scripts. Combine with --budget and --max-steps to bound what it can spend, and -y to approve gated actions for that run.

A GitHub repository

bash

export GITHUB_TOKEN=$(gh auth token)
devstation repo owner/name "add a Usage section to the README"

It clones the repository into .devstation/repos/, does the work, and finishes by proposing a pull request that you approve. The token needs push rights to the repository.

Watch, resume, list

bash

devstation status -f                                  # follow the live run from another terminal
devstation resume                                     # the last run in this workspace
devstation resume <id> "actually, use the existing helper"
devstation sessions                                   # every run in this workspace

Modes and shortcuts

Key	Does
/	Show every command under the prompt. Keep typing to narrow the list; Tab completes.
Shift+Tab	Switch mode: normal, then auto mode, then plan mode, then back. The rule above the prompt shows which is on.
↑ / ↓ and Enter	Choose an answer when the agent asks for permission or a plan is ready.
Ctrl-C	Interrupt the turn in progress.

Mode	What happens
Normal	Ordinary work runs; anything that changes things outside it asks first.
Auto mode	Actions run without asking. Critical ones still ask. The same as /approve on.
Plan mode	It reads, searches and looks things up, changes nothing, and ends with a plan to approve. The same as /plan on.

Plans

For work with several steps the agent keeps a checklist, shown as it goes: ✔ done, ◐ in progress, □ still to do. In plan mode it ends with that checklist and asks whether to carry it out; choose Yes and it starts, choose No and it keeps planning.

Slash commands

Command	What it does
/new	Start a fresh conversation in the same workspace. Also /clear.
/resume [id]	Pick up an earlier conversation, from a numbered list.
/rename <title>	Name this conversation.
/archive [id]	Hide a conversation from the list. /resume <id> still opens it.
/delete [id]	Delete a conversation for good, after asking.
/sessions	List conversations in this workspace.
/model [name]	Show the model, or switch it for this session.
/plan [on\|off]	Plan mode: read, search and propose; change nothing.
/approve [on\|off]	Run every gated action without asking, for this session.
/skill [name] [task]	List skills, or run one. Also /<name>.
/usage	Tokens and cost of this conversation. Also /cost.
/status /diff /undo /checkpoints	The same as the commands of those names.
/tools /mcp /memory /index /config	The same as the commands of those names.
/doctor /login /logout /upgrade	The same as the commands of those names.
/help /exit	Show the commands, or leave.

Plan mode

With /plan on, the agent may read files, search, run read-only commands and look things up on the web. Edits, installs, commits and anything with side effects are refused. It ends with a numbered plan and asks Carry out this plan?. Answer y and plan mode turns off and the work starts.

Tip

Use plan mode for anything larger than a small fix. Reading the plan first is the cheapest moment to change direction.

Skills

Skills are Markdown files of instructions you want reused:

text

.devstation/skills/<name>.md          this project
.devstation/skills/<name>/SKILL.md    the same, as a folder
~/.devstation/skills/<name>.md        every project

Skills in .claude/skills and ~/.claude/skills are read too, so skills written for Claude Code work unchanged. Optional frontmatter gives a name and a description. Run one with /<name> what to use it on.

markdown

---
name: release-notes
description: Draft release notes from the commits since the last tag
---
 
Read `git log` since the latest tag. Group changes into Added, Changed and Fixed.
Write for users, not for developers: say what changed for them, not which file.

Skills from the Marketplace

Agent skills bought on the DevStation Marketplace are ordinary skill files.
Put them in .devstation/skills/ for one project, or ~/.devstation/skills/ for all of them.