Skip to main content

Run and check tasks in Claude Code

After you connect Claude Code, you can start a saved ConsoleX task in your current session or add it as a Skill for repeated use. Use the folder that contains the files you want Claude Code to work with.

A real Claude Code session: /consolex:start runs the task, ConsoleX checks each delivery, and Claude asks before revising. Attempt 1 missed a rule, attempt 2 failed an AI check, attempt 3 passed; every attempt is in Run history. Example task and notes; music only, no voiceover.

Check the task type first​

This page runs saved tasks from either workflow: A uses the local main Skill; B loads the method and supported capabilities saved in ConsoleX. Both use the commands below; A requires its original main Skill on the computer. If your Claude Code workflow has not been saved as a task yet, follow Save a session as a task first.

Using ConsoleX Skills and tools in Claude Code​

For a task whose method is in ConsoleX, the plugin loads its saved instructions and brings along its configured capabilities:

  • Skills: exportable Skills are installed or refreshed on the computer. Some Skills cannot be exported, including a shared Agent's private Skills. A local Skill you own with the same name is preserved rather than overwritten; the plugin reports unavailable or skipped capabilities.
  • Tools and files: enable Tools and files for this computer under Settings → Coding agents. The plugin lets Claude Code call the task's supported ConsoleX tools; ConsoleX executes them with your configured connections and records their usage. Credentials stay in ConsoleX. Tools still waiting for a key or connection, and a shared Agent's tools, aren't available through this route.
  • Execution model: Claude Code uses its own model configuration. It doesn't switch to the model selected for the task's ConsoleX runs.

In an ordinary Claude Code session, local tools and Skills remain available alongside the task's capabilities. The plugin tells the agent to use the configured ConsoleX tools for the steps that need them and to report access problems before substituting its own. Use an isolated session when you want the task's Skills and ConsoleX tools prepared separately from your ordinary session. Claude Code still provides the basic file and Skill tools needed to work.

Start a task in the current session​

  1. List your tasks, optionally filtering by words from the name:

    /consolex:tasks
  2. Start one by name or by pasting its full ConsoleX task link:

    /consolex:start Weekly research brief
  3. To add instructions for this run, put them after --:

    /consolex:start Weekly research brief -- Use the notes in this folder for this week's brief.

/consolex:start loads the task and starts work in this session without installing a reusable task Skill. If several tasks match, choose the intended one using more of its name or its full link. Instructions after -- apply to this run; edit the task in ConsoleX to change future runs.

Add a task for repeated use​

Run:

/consolex:add Weekly research brief

The plugin installs a task Skill and reports its actual /cx-… name. Invoke that name in Claude Code to do the task, with any materials or request for this run. You can also copy the add command from the task's Run in Claude Code… dialog.

Adding or refreshing the task doesn't run it. To remove its local task Skill, use /consolex:remove followed by the task name or Skill name. The task and its history stay in ConsoleX. Scheduled tasks are managed separately.

Use an isolated session​

To get a terminal command for a new session prepared with the task's Skills, files and ConsoleX tools, run:

/consolex:run Weekly research brief

Review and run the returned command in your terminal. Asking for it doesn't start the work. Use /consolex:start when you want the work done in your current session.

If the task's method is a local Skill, that Skill must be available on this computer. A missing method Skill stops the task; the plugin tells you which one is needed.

After a check​

With automatic checks enabled, the plugin submits each delivery to ConsoleX and shows the verdict and run link. Open Run history in ConsoleX to read the result and check report alongside other attempts. See Read run results and checks.

  • Passed: the delivery met the checks within the evidence available. Review any remaining items that need your confirmation.
  • Failed: read which requirements were missed. By default, the agent reports them and waits for your decision.
  • Needs review or the check could not finish: review the evidence or check again as appropriate. Missing evidence doesn't establish that the delivered content is wrong.

Automatic revision is off by default. If you enable it in the task's checks settings, required output failures can trigger revision and another check, up to your selected limit of 0–10 rounds. Each revision and AI check has its usual cost. Items needing review and external actions don't automatically become rewrite requests.

For an active task session, /consolex:check checks the latest delivery, and /consolex:check off turns checking off for this session.

What the checks can see​

The plugin sends the delivery and available tool records from the task session. Recognized credentials are masked, but this isn't a guarantee that every sensitive value will be detected. Tool records are limited and may omit earlier, external or unsupported activity; a passed check isn't proof of everything done on your computer.

To send deliveries without tool records, set CONSOLEX_TOOL_EVIDENCE=off in the environment that launches Claude Code. Checks that need those records may then need review.

For file deliveries, ask Claude Code to identify the actual readable deliverable. A path in a reply alone doesn't prove the file's contents were checked. Check results also don't undo actions already taken, such as publishing or sending; put approval requirements in the task's run rules.

Scheduled runs​

In the Claude desktop app, first add the task, then use /consolex:schedule with its name and the time you want. Scheduling requires the desktop app's scheduled-task tools; the terminal CLI alone doesn't provide them.

Keep instructions about the work, publishing and failed checks with the task. If it already runs on a schedule elsewhere, turn that schedule off before creating another. For setup, examples, permissions, schedule changes and troubleshooting, see Schedule ConsoleX tasks in Claude Code.