Cline workflows: repeatable scenarios in one command

If every week you ask Cline for the same thing ("bump the version, run the tests, update CHANGELOG"), there is no need to type that list again. Cline workflows turn such a sequence into a markdown file you call from the chat with one command. Below: where to keep the files, how to write them, and what one run costs.

An important caveat about the docs

Cline has changed a lot. In the current docs.cline.bot navigation as of October 18, 2026 there is no standalone Workflows page any more, while the configuration page still lists workflows/ folders as part of the settings layout. The description of invocation below comes from an archived version of the Workflows page. If your version behaves differently, type / in the chat and look at the list of available commands, and check the Cline documentation and release notes.

Next to it, skills have grown in the documentation: modular instructions that you call with a slash command by skill name. For new scenarios it is worth checking first whether a skill is a better fit. The difference is covered below.

What a workflow is

A Cline workflow is a plain markdown file with a .md extension in which the steps are described in text. Cline reads the file and performs the steps in order using its tools: reading and writing files, running commands, asking you questions, plus any utilities installed on the system (git, gh, npm, docker) and connected MCP servers.

According to the archived documentation, you launch a workflow by typing a slash and the file name in the chat, for example /release.md.

Where to keep the files

According to the Cline configuration page:

• in a project: .clinerules/workflows/ (the older layout) or .cline/workflows/; these files can be committed and shared with a team;
• globally: ~/.cline/workflows/; Cline also looks in ~/Documents/Cline/Workflows/ (on Windows, Documents\Cline\Workflows).

If there is no workflows folder, create it. You can manage the list (enable, disable, edit, delete) in the rules and workflows panel inside the extension, but the button names depend on the version.

Example: releasing a library

File .clinerules/workflows/release.md:

## Library release

Go through the steps in order. After each step, report the result briefly.
If a step fails, stop and ask me what to do.

1. Make sure the working tree is clean: run `git status --short`.
   If there are uncommitted changes, stop.
2. Ask me which version we are releasing (patch, minor or major).
3. Bump the version in `package.json` and in `CHANGELOG.md`.
4. Run `npm test` and `npm run build`.
5. Collect commits since the last tag: `git log --oneline $(git describe --tags --abbrev=0)..HEAD`
   and add them to the new version section of `CHANGELOG.md`.
6. Create a commit `release: <version>` and a tag. Do not run `git push`:
   show me the command and I will run it myself.

Note two techniques. First, explicit stopping points: the agent asks for the version and does not continue after an error. Second, the irreversible action (push) is left to a person. For scenarios involving publishing, deletion or migrations this is the safest approach.

Example: a daily changelog

You do not have to spell out tool calls. The archived documentation stresses that Cline understands high-level instructions:

## Daily changelog

1. Show my commits from today: `git log --since=midnight --oneline --author="$(git config user.name)"`.
2. Ask what else I did that did not make it into commits.
3. Put it together into a short five-point summary and save it to `notes/<today's date>.md`.

Workflows, rules and skills: which to pick

Rules answer "how to behave always", a workflow answers "what to do now". If a scenario should be started by you and, when the task calls for it, by the agent, a skill with a good description fits. Keep rules separate from scenarios; the general Cline setup is described in Cline: connecting with your own key↗.

What one run costs

A workflow is an ordinary agent task, so you pay for model tokens. Let us estimate the release scenario: 12 requests, on average 30,000 input tokens and 800 output tokens each. Total input 12 × 30,000 = 360,000, output 12 × 800 = 9,600 tokens. Prices as of October 18, 2026, ignoring cache.

• claude-sonnet-5-5 (input $0.4, output $2 per 1M): 0.36 × 0.4 + 0.0096 × 2 = 0.144 + 0.0192 = $0.1632.
• gpt-6.1-sol ($0.27 / $1.32): 0.36 × 0.27 + 0.0096 × 1.32 = 0.0972 + 0.012672 = $0.1099.
• deepseek-v4.1-flash ($0.013 / $0.038): 0.36 × 0.013 + 0.0096 × 0.038 = 0.00468 + 0.000365 = $0.00505.

For mechanical scenarios like this it is reasonable to start with a cheap model and move to a stronger one only if steps go wrong. Test on your own tasks: the intelligence index (39 for deepseek-v4.1-flash, 48 for claude-sonnet-5-5) does not guarantee the outcome on a particular scenario. In the OpenAI format Claude input is billed at the full input price; the discount for reading cache comes with the Anthropic format.

Current prices are available through GET https://api.zukko.pro/v1/models.

Connecting

You need a model to run it. In Cline choose the OpenAI Compatible provider, Base URL https://api.zukko.pro/v1, your fetch_... key from the dashboard↗ and a model id. Step by step this is in the VS Code extensions docs↗. So that automated scenarios do not overspend, set a daily limit on the key, as in the article on key limits↗.

FAQ

The workflow does not appear in the command list, what now?

Check the folder (.clinerules/workflows/ or .cline/workflows/), the .md extension, and that the file is not switched off in the panel. After adding a file you may need to reload the editor window. If workflows are not supported in your Cline version, write the scenario as a skill.

Can one workflow call another?

The documentation we checked does not describe this. It is safer to keep one scenario per file and run them one after another by hand.

Is it safe to put commands in a workflow?

Commands go through the same approval rules as ordinary Cline actions. Do not enable full auto-approve for scenarios that publish or delete, and leave irreversible steps to a person.

Do workflows need special models?

No, it is an ordinary agent task. Whether a stronger model handles your scenario better, check in practice: do two or three runs on different models and compare the result and the cost.

To try it, create a key in the Zukko dashboard, connect it in Cline and write a first scenario of three or four steps.