.claudeignore

[iuriguilherme]

Open issue on claude-code: anthropics/claude-code#16704

Agents use .gitignore to prevent themselves from accessing some files, which in my opinion is undesired behavior because that file was made for and serves the CVS.

Agents should have their own list of files to be ignored which may and should be different from the cvs ignore file.

[Vasanthdev2004]

This is a solid idea.

Using .gitignore as the effective agent visibility boundary is convenient, but it does mix two different concerns:

• version-control hygiene
• agent/tool access policy

Those do not always match.

A separate .claudeignore / .openclaudeignore-style layer would give users much better control over:

• secrets or sensitive files they do not want the agent to read
• large/generated files they want excluded from context even if they are committed
• cases where .gitignore is broad for build reasons but too broad for agent workflows

The maintainer question is mostly design scope:

• precedence vs .gitignore
• whether it applies to read only, or also search/glob/context collection
• whether it is merged from user + project + workspace scopes

But as a product direction, I think the motivation here is very reasonable.

[iuriguilherme]

This is Gemini CLI right now