Start free trial
Guides / Working with agents

Guides

Working with agents

Use Wiele from Claude Code, Codex, Cursor and other agents that run terminal commands, with the bundled skills and the setup prompt.

Any agent that can run terminal commands can use Wiele through the wiele CLI: Claude Code, Codex, Cursor's agent and others. The agent acts as the person whose session it uses. It sees what that person can see and can change what they can change. Agents don't count as members on your plan.

Agents without a shell, or clients that prefer remote tools, can use the MCP server instead.

Set up with one prompt

The quickstart prompt installs the CLI, signs you in, creates an organization and a workspace, and saves a first folder. Paste it into the agent's chat. The agent asks you for your email, the sign-in code and to add a card, and runs everything else.

Install the skills

The CLI ships three agent skills, Markdown instructions that teach an agent the safe way to use Wiele:

Skill What it covers
wiele-workspace Find, fetch, read, edit and push files; handle stale heads, conflicts and payment errors; restore versions; retention.
wiele-conflicts Review a merge plan and resolve whole-file conflicts against a pinned draft.
wiele-sync Turn live uploads and download sync on and off without losing local work.

wiele skills install writes them into a new folder as NAME/SKILL.md:

wiele skills install --directory /tmp/wiele-skills

The folder must not exist yet, so install into a fresh folder and move the three skill folders to where your agent reads skills. For Claude Code, that's ~/.claude/skills for every project, or .claude/skills in one project:

mkdir -p ~/.claude/skills
mv /tmp/wiele-skills/wiele-* ~/.claude/skills/

For Codex, Cursor and other tools, use the folder their docs name for skills or rules. If a tool has no skills, print a skill and add it to the project's AGENTS.md or rules file:

wiele skills print wiele-workspace

The skills match the CLI version that wrote them. Install them again after you update the CLI.

Sign the agent in

If you can, sign in yourself in a terminal with wiele login before the agent starts. The agent then uses the saved session.

If the agent has to drive sign-in, it runs wiele auth start --email ADDRESS --json, asks you for the code, and pipes it into wiele auth verify --challenge ID --code-stdin. See Authentication.

Cloud sandboxes and containers usually have no keychain. Use a session file there:

mkdir -m 700 ~/.wiele
wiele login --credential-file ~/.wiele/session.json

Tips for agent runs

  • Pass --json --non-interactive when the agent reads results. Check ok and the exit code, and follow nextActions on errors.
  • Small text files can be read and written without a checkout: wiele read PATH and wiele write PATH --from FILE --if-sha256 HASH. Every write needs a precondition, so an agent can't overwrite a change it hasn't seen.
  • For larger work, fetch the folder, let the agent edit the files with its own tools, then wiele status, wiele diff and wiele push -m "MESSAGE".
  • To search content, fetch the folder and search it locally with rg or grep. Server-side search is off.
  • Give an agent its own copy for bigger changes, so a person reviews and merges the result.
  • Ask the agent to report the revision SHA after a push. A push isn't saved until the command reports a committed revision.

Example requests

Once the skills are installed, plain requests work:

  • "Put ~/projects/proposal in Wiele, in the client-work workspace."
  • "Fetch client-work:/acme and summarize what changed this week."
  • "Make a copy called acme-rewrite, rewrite the proposal there, and show me the diff against main."
  • "Restore budget.xlsx as it was before yesterday's push."
  • "Keep ~/Wiele/notes in sync while I work."