Swap heavy tool connectors for light CLI skill files
Idea + starter prompt, not run by us yet
For: Developers whose agent burns tokens on Jira or GitLab connectors
The Jira and GitLab connectors were eating tokens, so he installed their CLIs instead. He had Claude run --help for each CLI and its subcommands, saving every available action in its own Markdown file. Those became two small plugins, so the agent stops relearning the commands each time. He says they use far fewer tokens but gives no numbers.
Steps
- Install the official CLI for the tool and log in yourself in your terminal.
- Ask Claude Code to walk the CLI's help pages and save each command as a Markdown file.
- Group the files into a skill or plugin the agent can load.
- Test a few read-only commands before letting it run anything that changes data.
- Keep using MCP where you need its extra features, such as progress updates.
Starter prompt
I use [tool name] through its CLI [cli name], already logged in. Run the help command for the CLI and for every subcommand. Save one Markdown file per action into [folder], with the command, what it does, its flags and one example. Read only. Do not run any command that creates, edits or deletes data. When done, list the files and tell me which actions change data so I can mark them as needing approval.
Original post: Hacker News · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.