Claude Code Reads AGENTS.md Only When Telemetry Enabled

A recent blog post from Szypowi’s development team announced a new capability for Anthropic’s Claude language model: the ability to read and interpret code within its own agents, but only when telemetry is enabled. The update, detailed in the post titled “Claude code reads agents – MD only when telemetry is on,” explains that the model can now ingest source files and analyze their structure, improving its performance on programming tasks that involve internal agent logic. The feature is gated behind telemetry to ensure that user data is not inadvertently exposed, allowing developers to opt‑in for the enhanced functionality while maintaining privacy controls.

The announcement has already sparked discussion on Hacker News, where the post received 120 upvotes and 44 comments. Community members praised the technical depth of the feature, noting that it could streamline debugging workflows and enable more sophisticated agent behaviors. Some commenters raised concerns about the reliance on telemetry, suggesting that clear opt‑in mechanisms and transparent data handling policies will be essential for broader adoption. The developers responded by outlining the telemetry settings and emphasizing that the code‑reading mode remains fully controllable through the model’s configuration.

Overall, the introduction of telemetry‑controlled code reading marks a notable step toward more autonomous and context‑aware agents in Claude. By allowing the model to access and analyze its own codebase only when explicitly permitted, the team balances functionality with user privacy. As developers experiment with this new feature, it will likely influence how future iterations of conversational agents are built and deployed in production environments.