#Using Arachne well
The difference between a good Arachne run and a great one is how you set it up: describe the outcome, hand over what you know, choose the right amount of freedom, and steer the first run.
#Overview
Arachne is at its best when you treat it as a capable pair of hands working toward a goal you define. You do not script the steps — you describe the result you want, give it what you already have, and let it find the path across real websites and applications. The clearer your outcome and the better your starting materials, the better the work that comes back.
This page is a practical guide to getting strong results consistently. The habits here are simple, they compound, and they apply to everything Arachne does — browsing, research, building, forms, and creative work. None of them require you to understand how Arachne is built; they are about how you ask and how you stay in control.
#What you can do
To get the most out of Arachne, you have a handful of levers entirely under your control:
- Describe the outcome, not the steps. Say what "done" looks like and let Arachne work out how to get there.
- Hand over what you already know. Links, constraints, examples, and preferences all sharpen the result.
- Choose a permission level that fits the task. Stay close for unfamiliar or sensitive work; give more freedom for long, well-understood jobs.
- Watch the first run. Seeing the live work lets you steer early, before a small misunderstanding becomes a long one.
- Iterate in plain language. When something is almost right, ask for the change in words — no need to start over.
- Reuse what works. When a task is something you will want again, turn it into an automation or save it as a skill.
#How to use it
#1. State the outcome clearly
Lead with the result, not the procedure. Arachne finds the path; your job is to define the destination.
- Name the deliverable: a comparison, a report, a working site, a gathered set of sources, a finished form.
- Say what "good" means to you — the things you care about, the format you expect, the level of depth.
- Include the constraints that matter: a budget, a deadline, a region, a tone, a style, things to avoid.
A vague goal forces Arachne to guess; a specific goal lets it aim. "Research the outcome I want" beats "go look around" every time.
#2. Give it what you already have
You almost always know something that will save effort. Hand it over up front:
- Links you already trust — a starting page, a source, an account to operate, an example to match.
- Constraints — anything that must be true of the result, and anything that must not happen.
- Examples — "make it like this" is one of the most useful things you can give.
- Preferences — your usual format, your language, the look you want.
Arachne remembers what matters about your work over time, so you repeat yourself less as you go — but for a fresh task, a few details at the start go a long way.
#3. Pick the right permission level
You decide how much freedom Arachne has, and you can change it at any moment. Match the level to the task:
| Level | Use it when |
|---|---|
| Look only | You want to inspect or gather without changing anything. |
| Plan first | You want to approve the approach before any action. |
| Ask before acting | The task touches something sensitive and you want a say on each change. |
| Work in your workspace | The work is contained to your project and you trust it to flow there. |
| Act automatically | The task is long, well understood, and you want it carried through on its own. |
Start cautious when you are exploring something new, and give more freedom once you trust the path. Whatever level you choose, stopping always works immediately.
#4. Watch the first run
The first run is where you learn how Arachne is interpreting your goal. Watch it live:
- Confirm it understood the outcome the way you meant it.
- Catch a wrong assumption early, while it is cheap to correct.
- Notice where it pauses for your permission so you can keep things moving.
Because every action is automatically checked before it is reported done, you can trust a "done" — but watching the first run is how you make sure "done" means what you wanted.
#5. Iterate in plain language
Most results are a conversation, not a single shot. When something is close:
- Ask for the specific change in words — "make the comparison include price," "shorten the report," "change the layout."
- Ask for a different format if another of the eight fits better — a chart instead of a table, a dashboard instead of a report.
- Keep refining until it is right; you are building on the work, not restarting it.
#6. Reuse what works
When you find yourself asking for the same kind of work again, stop re-explaining it:
- Turn a repeatable task into an automation so it can happen on demand.
- Save a way of doing something as a skill you invoke later by name.
Everything Arachne produces is kept as an artifact you can find, reopen, and reuse — so good results become a foundation, not a one-time output.
#Examples
Outcome-first request
Build me a one-page site for my cafe — menu, hours, and a contact form.
Calm look, warm colors. Here's a photo of the place to match the vibe.You see a working site you can open, looking the way you described, saved as an artifact you can reopen and ask to change in plain language.
Giving links and constraints
Compare these three tools for me on price, free tier, and support.
Here are their pages. I care most about the free tier.You see a comparison laid out side by side against exactly the things you named, with the free tier weighted the way you asked, and a source pack so you can check any claim.
Steering on the first run
Research the best options for a standing desk under a set budget,
and watch it work so I can adjust as it goes.You watch the live work, nudge it toward the brands you care about as it gathers, and receive a synthesized result with its sources gathered — shaped by the steering you gave early.
#Tips
Be specific about the result, never the steps. "A comparison of these three, ranked by value" gives Arachne a clear target; "click here, then there" fights against what it does best.
- Start at a cautious permission level for anything unfamiliar or sensitive, then raise it once you trust how the task is going. You can change the level mid-task.
- Always watch the first run of a new kind of task. It is the cheapest moment to correct a misunderstanding, and it teaches you how to phrase the next one.
- When a result is close but not right, refine it in plain language instead of starting over — and ask for a different one of the eight formats if another fits better.
- If you will want a task again, save it as a skill or turn it into an automation so routine work happens on demand without re-explaining it.