Anthropic · 2026-09-18 · major
Claude Code 2.1.277 — AGENTS.md works when there's no CLAUDE.md
Claude Code 2.1.277 reads AGENTS.md as its project instructions in any repo that has no CLAUDE.md, so one file can brief several coding agents. The release also retires the TaskOutput tool and strips invisible Unicode from prompts.
One AGENTS.md can now brief Claude Code too, in any repo that has no CLAUDE.md.
Quick facts
| Version | v2.1.277 |
|---|---|
| Maker | Anthropic |
| Released | 2026-09-18 |
| New file read | AGENTS.md, when no CLAUDE.md exists |
| Where to switch | "Project instructions" in /config |
| Not yet on | Bedrock, Vertex, Foundry |
| Removed | TaskOutput tool |
What is it?
AGENTS.md support arrives in Claude Code 2.1.277: in a project with no CLAUDE.md, the CLI reads AGENTS.md as its project instructions instead. Which file wins is a setting under "Project instructions" in /config. The support is not on Bedrock, Vertex or Foundry yet.
How does it work?
The lookup is a fallback rather than a merge — CLAUDE.md still takes priority wherever it exists, so nothing changes in a repo already set up for Claude Code. Start-up got faster for scripted use in the same release: the first turn of an SDK or headless -p session no longer waits on the per-directory CLAUDE.md lookup.
Why does it matter?
AGENTS.md is the instruction file other coding agents already read, so a repo can now brief several tools from one place instead of keeping parallel copies that drift apart. Two hardening changes ship with it. A subagent's result now reaches the main agent under a header that marks it as subagent output, with the body indented, so text inside a result cannot pass as the session's own instructions. Invisible Unicode formatting and tag characters are removed from prompts, and the cleaned prompt is shown for review before it is sent.
Who is it for?
teams running more than one coding agent in the same repo
Frequently asked questions
- Does Claude Code use AGENTS.md if the repo already has a CLAUDE.md?
- No. Claude Code 2.1.277 reads AGENTS.md only in a project with no CLAUDE.md, so an existing setup keeps working exactly as before. If you want to choose deliberately rather than rely on the fallback, the file is selectable under "Project instructions" in /config.
- Is AGENTS.md support available on Bedrock, Vertex or Foundry?
- Not yet. The Claude Code 2.1.277 notes flag AGENTS.md support as unavailable on Bedrock, Vertex and Foundry, so teams on those providers still need a CLAUDE.md. Two other changes in the release are provider-specific in the other direction: the Bash sandbox instructions and workflow-script agent prompts on those three platforms were reworded.
- What replaced the TaskOutput tool that Claude Code removed?
- Claude Code now reads a background task's output file with the ordinary Read tool, and the deprecated TaskOutput tool is gone in 2.1.277. Two knobs stop doing anything as a result: the taskOutputMaxChars setting and the TASK_MAX_OUTPUT_LENGTH environment variable no longer have any effect, so scripts that set them can drop them.
- Which Claude Code crashes does 2.1.277 fix?
- Claude Code 2.1.277 is heavy on crash fixes, several of them triggered by a malformed ~/.claude.json — a bad theme value crashed at launch, and a bad claudeAiMcpEverConnected value crashed /mcp and /plugin manage. It also fixes crashes when a prompt contains terminal colour codes, when resuming a session whose history holds an assistant message stored as a plain string, and on slow machines as the first spinner appeared.
Try it
npm i -g @anthropic-ai/claude-code@2.1.277