Claude Skills in Cowork: Use Repeatable Workflows

Claude Cowork is a collaborative workspace where tasks can be carried out with AI assistance. Skills bring repeatability to that process: instead of explaining the same procedure for every task, you invoke a skill that defines the steps and apply it to the task at hand. The skill handles the how; the task handles the what.

The challenge is keeping the approved procedure and the resulting work together. Tools like fenbs hold the skill content, the task record and the history of what was done, so the next person or assistant picking up the work starts from the full picture.

How skills fit into Cowork

In Cowork, each task has a goal and may have a specific way it should be done. A skill defines that way: the review steps for a document, the format for a summary, the questions to ask before a decision. Invoking the skill at the start of the task aligns the AI's approach with the team's agreed procedure, without anyone having to re-explain it.

For how Claude Cowork works as a product, see what is Claude Cowork. For how skills are structured as files, see claude skills files. For finding community skills to apply to Cowork tasks, see claude skills marketplace.

Choosing the right skill for a Cowork task

Match the skill to the task type, not to the subject. A document review skill works on any document — a contract, a spec, a report — because it defines how to review, not what to review for. A subject-specific skill (one that only applies to legal contracts, for example) is narrower but may be more accurate for that domain.

• Use a general skill when the procedure is the same regardless of subject: summarize, review, compare.
• Use a specific skill when the subject requires domain rules: a legal review, a code security check, a financial reconciliation.
• Avoid using multiple skills on the same task unless they address different stages: one for the initial review, one for the final check.

Applying a skill to a Cowork task

1. Open the Cowork task and read the goal before selecting a skill. The goal tells you what the task needs to produce; the skill tells you how to produce it.
2. Invoke the appropriate skill with the task input as the argument.
3. Review the output before accepting it. Check that the skill's steps were followed: if the skill says "list each gap as a separate item", verify that the output does that.
4. If the output is incomplete or off, check whether the input matched what the skill expected. Adjust the argument format before editing the skill.
5. Once the output is correct, record it in the task alongside the skill used. That way the next person can see both what was produced and how.

Checking that the skill ran correctly

A skill can be invoked successfully and still produce a wrong result. The invocation is not the check — the output is. After running a skill, compare the output to the skill's stated criteria. If the skill says "each action item must have a named owner", check that every action item has one.

For more on structured evaluation of skill outputs, see claude cowork examples. For comparing Cowork to Claude Code for development tasks, see claude cowork vs claude code.

When not to use a skill for a Cowork task

Not every Cowork task benefits from a skill. Tasks that are genuinely unique — writing a first-contact message for a new client relationship, synthesizing feedback from an unusual meeting format — often need a tailored approach that a reusable skill cannot provide without getting in the way.

Watch for tasks where you spend more time adapting the skill's output than you would have spent completing the task from scratch. If you regularly have to undo what the skill did before you can make progress, the skill is not suited to that task type. Note it in the skill record so the team does not run into the same friction.

A skill that is occasionally wrong is less useful than no skill at all, because the wrong output requires correction and the correction takes longer than a clean start. Reserve skills for task types where the procedure is well understood and the expected output is predictable. A consistent skill applied to the right task type produces consistently useful work.

How to do this with fenbs

1. Create a free account and add a project for the Cowork work you are tracking.
2. Add the approved skill text to Skills and prompts in fenbs. Include a note on which Cowork task types it is for and the argument format it expects.
3. For each Cowork task, create a corresponding task in fenbs. When the skill runs, record the output and the skill name in the task notes so the procedure is traceable.
4. Connect your AI app over MCP from Integrations. It reads the approved skills and the task history before starting, so it can follow the same procedure on the next task of the same type.