Agent Skills' simplify-ignore hook stops deleting the code you hid from Claude

Addy Osmani's Agent Skills, version 0.6.13, fixes two bugs in its simplify-ignore hook for Claude Code. The hook hides code you mark from the code simplify skill. One bug could delete that hidden code, and the other brought back a change you had reverted.

Claude sees each hidden block as a one-line placeholder. If an edit deleted that line, say while clearing dead comments, the hook saved the file without the block, and the code was gone for good. Now the hook rejects that edit and tells Claude which placeholder to put back.

There is a limit. If the placeholder is still missing when the session stops, the hook restores the original file. Claude's changes are kept as a recovered copy in the hook's cache, and you merge them back by hand.

The code simplify skill tells the agent to revert any simplification that fails the tests, and agents revert with git. The hook then restored its old backup, bringing back the rejected change, and showed Claude the hidden code. Now the revert stays, and the code stays hidden.

The test driven development skill now fires when you paste a function and ask for tests, or paste a bug without mentioning tests. On Claude Opus 5.5 in Claude Code, both went from never firing to firing every time. Asking to change a function's behavior still fires in only one run of six.

The /review and /test commands run their skill in your main session. The docs said they started the code reviewer and test engineer personas, and now they say what the commands do. Review findings now use the same four severity labels as the skill.

Two more code samples are fixed. The frontend skill's error state called a refetch function that was never defined, and a zod title check accepted a title made only of spaces. The rest of the changes are in the 0.6.13 release notes.