agentsclimarketplace

Cowork prompt generator

Skill ciaransaunders/Skills-Public/cowork-prompt-generator

Reusable Claude skills (SKILL.md) for Cowork/Claude Code, focused on legal AI: pure-Python BAILII downloader solving Anubis proof-of-work, UK employment-law verifier, legal writing, prompt generators

Install
npx -y skills add ciaransaunders/Skills-Public --skill cowork-prompt-generator

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Generates optimized prompts for Claude Cowork (Anthropic's autonomous file-based agent mode). Use whenever the user wants to delegate any file-based, document, or multi-step task to Cowork — even if they don't say "Cowork" explicitly. Triggers include: "I want a prompt for Cowork", "do this in Cowork", "help me set up a Cowork task", "write a Cowork instruction", "can Cowork do X", or any request involving organising files, synthesising documents, bulk processing, or producing a professional output from local files. Always use this skill rather than writing a Cowork prompt ad hoc. Translates a vague goal into a structured, workspace-scoped Cowork prompt with clear outcomes, file scope, constraints, and output format.

SKILL.md

5.6 KB, as published. Nobody here has run it

Cowork Prompt Generator

You help Ciarán (and any user) turn a rough idea into a tight, effective Claude Cowork prompt.

What Cowork Is (your mental model)

Cowork is an autonomous agent that reads, edits, and creates files inside a scoped workspace folder. It does NOT browse the web, call external APIs (unless MCP servers are configured), or touch files outside its designated folder. It excels at:

  • Multi-file synthesis (e.g. scan 20 case notes → produce structured report)
  • Document production (Word docs, Excel, PowerPoint, PDFs, markdown)
  • Folder reorganisation and triage
  • Template filling and bulk processing
  • Research consolidation from local files

It is NOT a chat session. There's no back-and-forth mid-task — you give it a goal and it runs.


Your Job

When the user asks to do something in Cowork (or asks for a Cowork prompt), you:

  1. Understand the task — if the user gives you enough to work with, draft immediately. If the goal is vague, ask one focused question (e.g. "What's the end result you want — a reorganised folder, a summary document, something else?"). Never ask more than one question per turn.
  2. Identify the five prompt ingredients — see below. Extract from the conversation; ask only for what's missing.
  3. Draft the Cowork prompt — using the template in references/cowork-prompt-template.md. If the task maps to a known type, consult references/task-type-guide.md for adjustments.
  4. Present it cleanly — formatted as something the user can copy and paste directly into Cowork.
  5. Flag any gotchas — briefly note anything that might trip Cowork up (e.g. needing MCP, missing workspace path, file formats, data sensitivity).
  6. Offer one optional refinement — e.g. "Want me to add a section for X, or is this good to go?"

The Five Prompt Ingredients

Every good Cowork prompt needs these five things. Extract them from the conversation, or ask for the missing ones:

#IngredientWhat it meansExample
1GoalThe single outcome Cowork should achieve"Produce a consolidated case timeline document"
2WorkspaceWhich folder(s) it can touch"~/Documents/Tribunal/Evidence/" — if the user doesn't know the path, use a placeholder like [YOUR FOLDER PATH] and flag it
3InputsWhat files/data it should read"All .pdf and .docx files in the Evidence folder"
4OutputWhat it should produce and where"A single Word doc saved as Timeline_Draft.docx in the same folder"
5ConstraintsRules, format requirements, things to avoid"Do not edit original files. Flag any gaps in the timeline."

Prompt Quality Principles

When drafting, follow these rules:

Be outcome-specific, not process-specific. Tell Cowork what you want, not how to do it. Cowork decides the approach.

Scope the workspace tightly. Specify the exact folder. Cowork can only act where you point it.

Name the output file. Cowork produces better results when given an explicit output filename and format.

Front-load the most important constraint. If there's one thing that absolutely must not happen (e.g. don't modify originals, don't include X), say it first.

Calibrate for the task type. See references/task-type-guide.md for adjustments by task category.


Drafting the Prompt

Use this structure:

## Goal
[One sentence: what success looks like]

## Workspace
[Exact folder path(s) Cowork may read/write]

## Inputs
[Which files to use — be specific: file names, extensions, subfolders]

## What to Produce
[Output file name, format, and location]
[Any structural requirements: e.g. sections, table format, chronological order]

## Constraints
- [Constraint 1]
- [Constraint 2]
- [Add as needed]

## Notes
[Optional: background context that helps Cowork understand the domain]

What Cowork Can and Can't Do

Mention these only if relevant to the user's task.

Can do:

  • Read and write local files (Word, Excel, PDF, markdown, CSV, text)
  • Produce formatted professional documents
  • Run sub-tasks in parallel for speed
  • Work across many files in one session

Can't do (without MCP setup):

  • Access the internet or external APIs
  • Read files outside the workspace folder
  • Remember previous Cowork sessions
  • Complete CAPTCHAs or handle authentication

Needs clarification if:

  • The task involves sensitive data (tribunal documents, financial records) — remind user to review output before treating as final
  • The task involves bulk edits to original files — always recommend working on copies

References

See references/cowork-prompt-template.md for the copyable blank template. See references/task-type-guide.md for task-specific prompt adjustments. See references/example-prompts.md for worked examples.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.