Start with a template or work it out in a chat, then save it as a reusable task. Check each run against your standards, and use what you learn to improve the task over time.
Start from a task template
All templatesNothing fits? Chat it through
An example in four parts. In a chat, someone asks for a brief on what three competitors shipped, then corrects it once: only the last 7 days, with an official source for every item. They save the chat as a task; AI turns the conversation into requirements that include the correction, with three suggested checks. The task’s first run comes back with a check report: two checks pass, and one needs review because Atlas’s price change only appears in a forum post. The report shows the evidence behind each verdict, the run’s usage and estimated cost, and what the checks cost. The owner reviews the result, marks it Fail and gives the reason: the price change has no official source. Then a suggestion drafted automatically from that run tightens the requirement: price changes need an official pricing page. Once applied, the next run passes all three checks.
What’s different
An answer you have to check yourself, and explain again next time, isn’t finished work.
| Chat assistant | ConsoleX | |
|---|---|---|
| Next time | Paste the prompt again and repeat last time’s corrections. | Your corrections are already in the task. Bring new material and run it. |
| Is it good enough? | You read it line by line and check it yourself. | Checked item by item against your standard, with evidence for every verdict. |
| What it can’t verify | Mixed in with everything else, for you to spot. | Listed separately for your call, never quietly passed. |
| When it misses | You work out what to change and rewrite the prompt yourself. | It becomes a suggested improvement to the task. Approve it, and later runs follow it. |
Use cases
Pick the work you already repeat and hand over. Each use case starts from ready-made templates with their checks, and you can change anything to fit how you work.
Turn this week’s interview, call or voice note into LinkedIn posts that sound like the client and use only what they said. When the client sends notes, the second version follows them.
Every run is checked for
Turn an article, video or episode into a post for each channel you use, and edit drafts to your own rules and the latest feedback.
Every run is checked for
Templates
A monthly SEO and AI-visibility report for each client, a recap or proposal from your notes, or a sourced brief before a decision.
Every run is checked for
Templates
Announce what’s new on every channel, turn what shipped into release notes and fill in directory listings, all from one product profile.
Every run is checked for
Profiles
A profile keeps what your work is for: a client’s voice, words and facts, or a product’s facts and the claims you can back up. Tasks that name it follow it on every run, and their checks read it.
Daniel Kim
Give AI a client’s approved posts or style guide, or your product’s web page or brief. It drafts the profile; you check it and save it.
Choose the profile in a task. Each run records the version it was written with, and its checks read that same version.
Update a preference in the profile, and every task that uses it follows it from the next run. Rules a run suggests from client notes go in only when you add them.
Claude Code
With the ConsoleX plugin, Claude does a task on your computer, in your own folders. When it hands over the result, ConsoleX checks it against the task’s standard and adds it to the same run history.
/cx-acme-release-notes v2.4
Bash(git log --oneline v2.3..v2.4)
⎿ 14 commits
Release notes · Acme Notes 2.4
ConsoleX ✗ attempt 1 failed: No user-facing change left out; waiting for your decision
One check didn’t pass: “Export to CSV” shipped in a41c9e2 but isn’t in the notes.
Revise the release notes for this item?
Added “Export to CSV” under Improvements.
ConsoleX ✓ delivery passed on attempt 2
Add a ConsoleX task to Claude Code, or save a Claude Code session as a task with /consolex:save.
A model from another family does the checking. Claude sees how to do the work, never your standard, and after a check only the items that didn’t pass.
When a check fails, Claude tells you and waits for your call, unless you’ve let it fix failed checks on its own.
Needs a paid plan.Set up the plugin
Run history
What passed, what you decided, what changed, what it cost.
Weekly competitor brief
Every Monday at 09:00
See which item failed, which needs your call, and why. A run you review counts as your Pass or Fail.
Every run in ConsoleX records the model, usage and estimated cost.
Run it when new material comes in, set a schedule, or do it in Claude Code. Every run is checked against the same standard.
How it improves
Failed checks, your reviews and your own requests become suggested improvements. Each one you approve is a new version.
Each run uses the task’s current version.
Each item gets a verdict and evidence. Review the run yourself when you want: Pass or Fail, with your reason.
Drafted from failed checks and the reasons you gave, with every edit marked.
Nothing changes until you do. Any earlier version can be restored.
Next run: new version, same standard
Before you start
It’s built for people who deliver the same kind of work again and again and answer for it themselves: freelancers, creators, consultants and founders running a one-person business. If you only ask AI the occasional question, a chat assistant is enough.
Working it out in a chat is the same. What’s different is what happens next: the chat becomes a task with your standard saved in it, so later runs are checked item by item, with evidence and cost, instead of starting over.
No. Most tasks work without one. Templates that write for a client or about your product ask you to choose one, and you can draft it with AI from a document or web page without leaving the task.
Yes, on a paid plan. Install the ConsoleX plugin and connect your computer once. Then add a task to Claude Code, or save a Claude Code session as a task. Claude does the work with its own model; ConsoleX checks each delivery and records it in the task’s run history.
Templates that write emails or posts only draft them. You review the result and send or publish it yourself. ConsoleX can email you when a run finishes.
No. Suggested improvements, including ones drafted automatically after a failed check, take effect only after you review and approve them, and every version of the task can be restored.
Rules that can be verified directly, such as required sections, links or banned words, are checked without a model. Where judgment is needed, the verdict has to cite evidence. If the evidence isn’t there, the item is marked for review rather than passed. You make the final decision.
No. A check compares the result with your standard. It doesn’t guarantee that every fact is correct. You can review any run yourself and mark it Pass or Fail; your review is what counts in pass rates.
Each run shows its usage and an estimated cost, with checks estimated separately. Estimates can be incomplete, don’t include charges from external services, and are not a spending cap.
Both. Run it whenever new material comes in. Once it works the way you want, you can also put it on a schedule.
Make it a task. Start from a template, and every run gets checked.
