untangle-codebase skill
A skill that runs a full refactor so agents read code more easily: it finds hidden links and needless dependencies.
- 1,149 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill edits many files and, with the extras selected, deletes comments and tests.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/untangle-codebaseDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill edits many files and, with the extras selected, deletes comments and tests.
This is third-party code. Review the repository files before installing.
What it does
untangle-codebase sets the agent a goal of finishing a refactor of the whole codebase in one pass. The point is that agents read less per change, break code silently less often and find root causes faster. The skill is started by hand, the model does not invoke it on its own, and it takes an optional argument for an area or concern. Work runs in four steps. First the agent asks whether you want extras: removing comments that only restate the code and removing tests that would not catch a real bug. Then it lists problems without changing anything, asking whether pieces should change together. There are three kinds of errors: hidden links (copies of logic, implicit call order), needless links (relying on others' internals, stable code depending on volatile code) and surplus (wrappers, unused extension points, dead code). Special cases are listed separately, such as an `if` for one value, a swallowed exception or a disabled test. Then the agent designs fixes from first principles, looking for one fix behind several problems, and carries them all out. The result is recorded in a decision log.
Who it is for. Developers who hand code to agents and want to put the structure in order first so that changes cost fewer tokens and break less.
Good fit when
- Agents on this project often break neighboring code because links between parts are not visible anywhere.
- You want to cut the tokens spent finding the right files.
- Copies of logic and wrappers piled up and need pulling into a common form.
- Tests and comments became noise and what catches no errors should go.
- Before a long stretch of agent work you want one review of the structure.
Not a fit when
- You need a targeted edit of one file, while the skill walks the whole codebase.
- The project has no tests and no version control, and a big refactor without a safety net is dangerous.
- The code is someone else's and cannot be changed.
Example request
/untangle-codebase focus on the data access layer. Extras: remove comments that restate the code.Limitations
The skill changes many files in one pass, so run it on a separate branch and check the result with tests and a diff review. It can delete comments and tests if you pick those extras, so read their description first. It is invoked only by hand and the model never starts it on its own. The repository holds one skill and no others. The result depends on project size and the model.
How to disable. Delete the skill folder from `~/.claude/skills/` or uninstall the `untangle-codebase@caldo-skills` plugin in Claude Code.
FAQ
When does the skill run?
Only by hand, for example with `/untangle-codebase`. The model does not invoke it automatically.
What counts as a problem?
Three kinds: hidden links, needless links and surplus, plus special cases like an `if` for one value.
Can I limit the scope?
Yes, through an optional argument naming an area or concern.
What extras does it offer?
Removing comments that restate code and tests that would not catch a real bug. The choice is yours.
Which agents fit?
Any that work with the skills CLI: Claude Code, Cursor, Codex and others.
Discussion
No questions or reviews yet.
Related
SkillsA cleanup skill for agent-written code: finds surplus tests, self-checks and fallbacks and deletes them, keeping behavior
A skill that finds redundant abstractions and duplicate state in code and removes them safely, one proved cut at a time.
The /improve skill: a strong model audits a repository and writes self-contained fix plans, and you hand the execution to a cheaper model
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
A skill that finds AI slop in code: duplicate helpers, empty abstractions, weak tests. It cleans one pattern per commit.
An autonomous iteration skill for agents: modify, verify, revert failed changes, and repeat the loop toward a target metric
README in short
The README briefly describes the repository as a Claude Code plugin marketplace from Caldo with one skill. Installs with `npx skills add`, MIT license.