Skills, Real Workflows & Working Safely
You can now hand Cowork a task and get finished work back. This lesson is about doing that repeatedly and safely — turning one-off wins into a system you trust.
Your highest-leverage workflows
The goal isn't to automate your job. It's to do the same quality work in a fraction of the time. Find the tasks in your week that are repetitive and involve writing, research, or analysis — those are your gold mines. A few that pay off immediately in Cowork:
Communication, from your real material
Read the thread in
client-emails/and draft a reply that declines their extra request without hurting the relationship. Save it asdraft-reply.mdso I can review before sending.
Research and synthesis
Pull the key points from the reports in this folder into a one-page brief for my manager — audience is non-technical, focus on decisions and risks. Save as
brief.docx.
Decisions, pressure-tested
Read
option-a.mdandoption-b.md, give me a structured comparison — pros, cons, key tradeoffs — and a recommendation. Then be my harshest critic: what's the strongest case against your pick?
Repeatable analysis
Every file in
survey-responses/is a batch of feedback. For each, produce a short report with: top 3 themes, sentiment, and one recommended action. Save them next to the originals with a-reportsuffix.
That last one is the tell: the same shape of task, run again and again. That's exactly what Skills are for.
Skills — teach Cowork your repeatable work
A Skill is a saved set of instructions for a task you do often, so you don't re-explain it every time. Instead of writing a paragraph of context and formatting rules on each run, you capture it once as a Skill, and then just point Cowork at new material and say "do the thing."
Think of the difference between telling a new temp how to format the weekly report every single week versus training one assistant once and just handing them the numbers thereafter. Skills are that training.
Good candidates for a Skill:
- A report or document you produce in the same format on a schedule
- A way you like research summarized (your sections, your tone, your length)
- A checklist you run over every draft before it ships
- A standard way you clean or label a recurring kind of data
You'll build Skills in real depth in Module 9 (for Claude Code), but the idea starts here: anything you've now explained to Cowork twice is a candidate to save once.
Persistent context, revisited
Remember from Lesson 3: the more Claude knows about you and your standards, the better the output — without you re-typing it. Two ways that context persists:
- In Chat, through Projects — instructions and reference files that apply to every conversation in them.
- In Cowork, your work also persists in the files and folders themselves: keep a
context.mdor a style guide in your working folder, and ask Cowork to follow it. Pair that with Skills and Cowork starts every task already knowing how you like things done.
Working safely — you stay in control
Cowork is powerful precisely because it can change real files and use real tools. That power deserves a few habits.
Approve deliberately
Cowork asks before it does anything significant — creating or editing a file, using a connector, deleting. Read what it's about to do before you approve. Auto-approving everything defeats the safety that's built in.
Scope its access
- Point Cowork at the specific folder a task needs — not your whole drive.
- Start new or risky work on a copy of your files until you trust the loop.
- Add only the connectors a task requires, and grant the least access that gets the job done.
Mind what's sensitive
- Check your company's AI-usage policy before pointing Cowork at work documents.
- Be thoughtful with financial records, personal data, and confidential contracts — anonymize or work on extracts when you can.
- On Team/Enterprise plans, admins can set connector permissions and usage controls centrally.
Verify the output
Claude can be confidently wrong, especially on recent facts or fiddly numbers. Treat its work like a sharp junior's: review it, spot-check the figures, and you're responsible for what ships.
The mantra: scope the folder, approve the actions, verify the result. Do those three and Cowork is both powerful and safe.
Putting it together
A mature Cowork workflow looks like this:
- A working folder with your material and a
context.mdof how you like things done. - A Skill (or a saved prompt) for each task you repeat.
- The connectors that task needs, and nothing more.
- A habit of reviewing actions and verifying output.
Set that up once and "do my weekly report," "summarize this folder," "clean this data" become one-line requests that produce finished, trustworthy work.
Summary
- Your best Cowork wins are repetitive writing/research/analysis tasks — run them on your own material.
- Skills capture a repeated task once so you stop re-explaining it; anything you've described twice is a candidate (you'll go deep on Skills in Module 9).
- Persistent context comes from Projects (Chat) and from keeping a
context.md/style guide in your Cowork folder. - Stay safe with three habits: scope the folder, approve actions deliberately, verify the output.
- You're the manager: Claude does the work; you direct it and own the result.