Specification Extension Request: Standardized ignore files Support in Agents.md Specification

[sanathusk]

Summary

Extend the Agents.md specification to include standardized handling of ignore files. This would introduce a single .agentignore / .aiignore, file that all agentic tools can comply with, replacing the need for multiple, tool-specific ignore files (e.g., .clineignore, .claudeignore, etc.).

Motivation

Currently, each agentic tool tends to define its own ignore configuration, leading to fragmentation and duplication.
Examples:

• .clineignore for Cline
• .claudeignore for Claude
• Other tools introducing their own formats

This creates unnecessary friction for developers managing projects across multiple agentic ecosystems.

Proposed Solution

• Introduce .agentignore as the standard ignore file for agentic tools.
• Define in the Agents.md specification how this file should be interpreted (syntax, precedence, patterns).
• Encourage tool maintainers to adopt .agentignore as a unified convention.

Prior Art / References

• agentignore provides a practical reference for what such a standard could look like.
• Established conventions like .gitignore, .dockerignore, and .npmignore show the benefits of standardization in developer workflows.

Benefits

• Consistency: One ignore file works across tools.
• Simplicity: Reduces cognitive load for developers.
• Interoperability: Promotes cross-tool compatibility and collaboration.
• Ecosystem growth: Standardization encourages adoption and easier onboarding.

Next Steps

• Discuss inclusion of .agentignore in the official specification.
• Define the syntax and semantics clearly in Agents.md.
• Engage with tool developers to encourage early adoption.

[michelTho]

We are in favor of implementing this on Mistral Code if it is established as standard

[sanathusk]

@michelTho Mistral Code could lead this, more opensource agentics tools would follow, We could at least raise the PR for opensource once,once it gains enough trajection, more tools would adopt it, infact AGENTS.md also has similar origin story.

[rosmur]

Thanks for raising this! We need to prioritize this!

LLMs reading things that they shouldn't is a SERIOUS security and privacy risk that is going entirely overlooked.

For a common workflow using an LLM through Openrouter etc., there are at least 2 other parties that have access to your traffic. AI agents have show many times to read .env files so API keys and the like are sitting with cloud provides.