Claude Code: AGENTS.md Support Gated by Telemetry in v2.1.277

Claude Code v2.1.277 silently failed to load AGENTS.md project instructions when telemetry was disabled. This occurred because the feature was gated behind a remote feature flag that required a network connection to resolve, effectively disabling the local file reading capability for users with DISABLE_TELEMETRY=1 or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 set.

Remote Feature Flags Gating Local File Access

In version 2.1.277, the agents-md plugin was configured with isOnByDefault: false. The plugin's availability was determined by a remote feature flag called tengu_agents_md_mod. If Claude Code could not fetch this flag from Anthropic's servers—which happens when telemetry or non-essential traffic is disabled—the plugin remained unavailable and the local AGENTS.md file was never read.

This behavior was particularly impactful for users on third-party gateways like Bedrock and Vertex, where network policies often block the traffic required to resolve these flags, causing the feature to be missing entirely for those users.

Impact and Silent Failure

Users experienced a silent failure where the model would answer without the project instructions contained in AGENTS.md without any warning or error message. This led some users to believe the model was simply ignoring their instructions, potentially wasting time on prompt engineering when the file was never actually loaded into the context.

Workarounds and Configuration

Until the fix in v2.1.281, users employed several workarounds to maintain privacy settings while using project instructions:

  • The @path Import: Creating a one-line CLAUDE.md file containing @AGENTS.md allows the tool to load the instructions without relying on the feature flag.
  • Session-level Overrides: Using the --settings flag to clear telemetry environment variables for a specific session: claude --settings '{"env":{"DISABLE_TELEMETRY":"","CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC":""}}'.
  • Symlinking: Creating a symbolic link from AGENTS.md to CLAUDE.md (ln -s AGENTS.md CLAUDE.md).

Developer Response and Resolution

An Anthropic engineer (@mpoteat) confirmed the issue as a "human error" related to the implementation of a kill-switch for the rollout. The issue was resolved in version 2.1.281, which removed the telemetry dependency for reading local files.

"Sorry folks, this is a rollout artifact, we needed a way to turn this off remotely via feature flags if it broke something, and with telemetry off you don't get those. It's already been fixed as part of v2.1.281 releasing today."

Community Insights and Design Critiques

Technical discussions on Hacker News highlighted several design concerns regarding the use of feature flags for local behavior:

  • Privacy-Feature Trade-off: Critics argued that a privacy setting (disabling telemetry) should not silently disable unrelated local functionality.
  • Vibe-coding: Some users characterized this as "vibe-coding," suggesting that AI-generated patches to a codebase may lead to subtle, severe bugs that bypass traditional testing.
  • Standardization: There is a persistent community request for Claude Code to fully adopt the AGENTS.md and .agents/ standard over the proprietary CLAUDE.md format.
  • Global Configuration: Users noted a lack of support for a global ~/.codex/AGENTS.md or a user-level instructions file that persists across projects without manual copying.

Sources

Related

  • Dispatch
    Claude Code adds AGENTS.md support

    Claude Code now reads AGENTS.md when there is no CLAUDE.md, simplifying project setup and aligning with emerging standards.

  • Dispatch
    Claude Code and the AGENTS.md Standard: Interoperability Conflict

    A community-led push for Claude Code to support the open AGENTS.md standard for AI agent instructions has led to significant developer backlash after Anthropic closed the feature request without native implementation.

  • Project
    kristianvast/hermes-claude-auth

    hermes‑claude‑auth is a Python runtime patch that lets the hermes‑agent use a Claude Code subscription after Anthropic added OAuth validation. It installs a .pth shim that monkey‑patches the agent’s request builder, adds the required billing header, fixes prompt layout, and implements auto‑wait for subscription‑window rate limits. The installer works on Linux/macOS and Windows, provides uninstall scripts, and includes git‑hook + cron recovery to survive hermes updates.

  • Dispatch
    Claude Code 2.1.196 Prompt Steganography Analysis

    Claude Code version 2.1.196 uses prompt steganography to covertly mark requests based on user timezone and API base URL to detect resellers and model distillation attacks.

  • Dispatch
    Claude Code Best Practices Guide

    Anthropic provides a comprehensive set of best practices for optimizing Claude Code, an agentic coding environment, focusing on context window management and autonomous verification.