Writing Better Prompts

The fastest way to improve Alfrada OS outputs is to improve how you frame the task.

If you want copy-ready examples by job, start with Prompts By Industry; by tool, open the Tool Prompt Library.

The Four-Part Formula

Strong prompts usually include:

1. Objective: what success looks like
2. Inputs: files, sources, accounts, or assumptions to use
3. Output: exact format, audience, and level of polish
4. Guardrails: constraints, risks, tone, and review standards

Prompting Patterns That Work

For analysis

Analysis pattern

Analyze [subject] using [sources/files]. Return the answer as [format]. Highlight assumptions, risks, and where the evidence is weak.

This encourages evidence-backed output instead of generic commentary.

For creation

Creation pattern

Create [deliverable] for [audience]. Use [inputs]. Match this tone: [tone]. Keep it within [constraints]. Give me a draft first, then suggest the highest-value improvements.

This creates better drafts and makes revision easier.

For automation

Automation pattern

Set up a repeatable workflow that checks [sources] on [cadence] and produces [output]. Escalate only if [trigger] happens.

Automation needs triggers, cadence, and output expectations to be explicit.

For coding

Coding pattern

Help me with [engineering task] in [repo/codebase]. First inspect the relevant files, explain the plan briefly, then implement or propose the change. Call out risks, testing needs, and edge cases.

This improves reasoning before execution and lowers the chance of low-quality blind edits.

The Highest-Leverage Control: Effort

Before you polish another word of your prompt, check the Effort setting — it changes cost and depth more than prompt wording does. You'll find it in the chat input → model selector → Effort row. It controls how thoroughly Alfrada OS double-checks its own work, separate from which model you choose:

• Auto — Alfrada OS picks the best model and effort for each turn.
• Fast — no self-review; the fastest way to an answer.
• Medium — one self-review-and-revise pass before you see the result.
• Beast — reviews and revises until the work passes (capped), and on Auto also grades any charts or images it produces.

You can pin Effort from the selector, or just ask for it in chat:

Pin Fast for quick back-and-forth

Pin Fast — short answers only for the next few questions.

Quick triage does not need a self-review pass; Fast keeps answers short and cheap.

Switch to Beast for deep work

Switch to Beast and write a full competitive brief with sources for each claim.

Beast keeps reviewing and revising its own work until it passes — what a high-stakes deliverable deserves.

Picking A Model

Leave the model on Auto for everyday work — Alfrada OS routes each turn to a model that fits it. When you do want to choose, the model selector groups models into sections and opens on a curated Recommended list, so a good manual pick is never far away. For hard coding or agent work, pin Claude Sonnet 5 - Reasoning.

One exception: in Swarm mode your model pick doesn't apply — the Swarm assigns each worker its own model, and your pinned model and Effort take over again when the session returns to Agent mode.

When To Ask For The Swarm

Some jobs are really several independent jobs — research five competitors and merge the findings, or draft three sections that don't depend on each other. That is Swarm territory: a planner splits the work across parallel workers instead of one agent doing everything in sequence. Switch yourself with chat input → model selector → Mode row → Swarm, or just describe the job — when Alfrada OS thinks the Swarm would do it better, it pauses and asks with a consent banner, and nothing starts spending until you approve the plan. The Swarm page covers the full flow.

Invite the Swarm

This splits cleanly: research each of these five competitors independently, then merge everything into one comparison brief. Use the Swarm if that gets it done better.

Naming the independent pieces gives the planner an obvious decomposition, and inviting the Swarm up front means the consent banner confirms your plan instead of surprising you.

Memory And Continuity

Alfrada OS keeps long-term memories across sessions — it can search them, save new ones, update ones that changed, and delete ones you'd rather it forgot — and it can search your full session history whenever you ask. Corrected facts are superseded, not erased: when something changes, the old version stays in that fact's history with the dates it was true — see the Memory page for how to browse it.

Recall often happens on its own, but the reliable pattern is still principle 3 below — say it out loud:

Carry earlier work forward

Search history for the pricing analysis we did on [project] and carry the assumptions forward into this draft.

History search runs on request — asking explicitly is the reliable way to reconnect past work to the current session.

Attaching Files

Use the + button in the chat input to attach files — its tooltip reads "Upload files to include them in your chat." Whenever a playbook says "upload X first", this button is what it means. Uploads land in the session and show up in the Work Panel — the side panel that collects everything a session takes in and produces; see Files and Outputs for the full tour of its tabs.

Expect Questions Back

Alfrada OS has a built-in Ask User step: instead of guessing, it can pause mid-task and show you a short form — sometimes with options to pick from or a required field to fill in. A good prompt states where you would rather be asked than guessed: "ask me before spending money", "check with me before sending anything". See When Alfrada OS asks you questions for how the form works and where the auto-skip setting lives.

Five Principles For Power Users

These apply to every prompt, every session, every domain.

1. Provide constraints, not magic — Give the system explicit parameters, context, and boundaries to operate within. The more specific the frame, the better the output.
2. Name your deliverables — Always specify the exact output format you need: Markdown memo, structured CSV, 10-slide presentation. Never leave it vague.
3. Ask for continuity explicitly — When the job started earlier, tell Alfrada OS to search history and pull the relevant context forward.
4. Separate chat from outputs — Keep the chat channel strictly for steering and revision. Let the Work Panel handle the finished files.
5. Codify your success — When you nail a complex workflow, save it as a Playbook so you never have to design it from scratch again.

Common Mistakes

• asking for "something good" without saying what good means
• mixing multiple unrelated jobs into one prompt
• forgetting to define the audience
• not naming the desired file or output format
• forcing long answers when you actually want a decision or a shortlist

Copyable Prompt Packs

• Prompts By Industry for copy-ready prompts organised by the job your team does
• Tool Prompt Library for searchable prompts across research, coding, finance, content, social, and travel workflows
• use site search to jump straight to a tool name like Fast Browser, Code Execution, or Presentation Create
• replace bracketed placeholders like [company], [URL], and [audience] before running the prompt