Claude Code adopts AGENTS.md as Copilot reviews and six models retire Oct 19

Saturday is an interoperability and cleanup morning, so the useful work is how these four moves fit together. Anthropic stops making dual-agent repos maintain a CLAUDE.md symlink just to share AGENTS.md, GitHub makes Copilot review progress readable across commits, Copilot admins get a hard Oct 19 model cutover list, and Claude Code pins a no-charge auto-mode classifier path for API and enterprise fleets.

Treat today as a share-AGENTS.md, re-review, migrate-models, and pin-v2.1.278 day. Drop or keep CLAUDE.md deliberately on one multi-agent repo, request a second Copilot review after fixing open findings, inventory which deprecated models your policies still allow, and confirm `/status` shows Auto mode server when you expect no-charge checks.

Top Stories

Claude Code v2.1.277: AGENTS.md fallback when CLAUDE.md is absent Practical dev impact: Starting in v2.1.277 (Sep 18), Claude Code reads `AGENTS.md` when a project has no `CLAUDE.md`, so repos already documented for Codex, Cursor, Gemini CLI, and other AGENTS.md adopters stop needing a second instruction file or a symlink workaround. CLAUDE.md still wins when present, and you can change the behavior under Project instructions in `/config`. The fallback is not available yet on Bedrock, Vertex, or Foundry. The same release also adds `CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1` and optional static `headers:` on Claude apps gateway upstreams for forward-proxy egress setups. Pin past Friday’s skill-sync line into v2.1.277 (or straight to v2.1.278 below) on one shared monorepo that already has AGENTS.md and no CLAUDE.md, then confirm Claude actually loads the open format before you delete any legacy links.

Copilot code review: clearer progress, smarter resolve, batch commit messages Practical dev impact: As of Sep 18, Copilot code review’s overview comment is generally available with a current PR assessment, the review effort level used, and findings grouped into Open (including `new` for issues introduced by a later commit), Resolved since last review, and Previously missed. Comment titles summarize each finding so you can triage before diving into inline threads. Auto-resolution now honors replies that intentionally leave an issue open, and it can resolve with reasons such as `Won’t Fix` or `Incorrect` based on later commits. When you commit an eligible complete batch of Copilot suggestions (batches may also include non-Copilot comments), Copilot generates a relevant commit title and optional description. Re-request review after you address Open items and treat Previously missed as new signal, not noise from the first pass.

Copilot mid-October model deprecations: cut over by Oct 19 Practical dev impact: GitHub will remove Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 from every Copilot surface (chat, inline edits, ask and agent modes, completions) on October 19, 2026. Suggested replacements are Gemini 3.8 Flash, GPT-5.6 Sol (for GPT-5.5 and GPT-5.4), GPT-5.6 Luna (for the mini models), and Grok 4.6. Under default model enablement, Copilot Business and Enterprise get the alternatives automatically unless admins turned off the global default or disabled the replacement model; otherwise enable them via model policies. Inventory prompts, custom agents, and CI that hard-code the retiring ids, then verify the replacements appear in the model selector before the cutover rather than on the morning of the 19th.

Claude Code v2.1.278: server-side auto mode classifier by default Practical dev impact: Published early Sep 19 ET, v2.1.278 changes auto mode for Claude API and Enterprise users (and on Bedrock, Vertex, Foundry, and gateways) to default to the server-side classifier, which does not charge for classifier overhead when the server performs the checks. `/status` gains an Auto mode server row so you can see whether this session’s checks run on the server. Set `CLAUDE_CODE_AUTO_MODE_SERVER=0` to opt out on Bedrock, Vertex, Foundry, and gateways if your proxy cannot pass the safeguards fields through; the release warns on billed fallback. Fleet operators who already moved to v2.1.276 for the gateway fix should land on v2.1.278 today so AGENTS.md support and no-charge auto-mode checks ship together, then read the auto-mode classifier billing doc if a gateway notice appears.

Practical Impact Analysis

Today’s through-line is reducing dual maintenance while tightening the feedback loop around agents and models. AGENTS.md in Claude Code means one instruction file can serve mixed Claude and Codex or Cursor workflows, so teams stop drifting CLAUDE.md and AGENTS.md apart. Copilot’s review overview turns a wall of comments into Open versus Resolved versus Previously missed, which pairs cleanly with the mid-October model list, because you need the same admin discipline for review enablement and for which chat and agent models remain after Oct 19. Claude Code’s v2.1.278 auto-mode default then closes the weekend loop for API and enterprise fleets that care about classifier cost and gateway compatibility.

If you maintain a repo used by more than one coding agent, Saturday is the day to try Claude with AGENTS.md only and decide whether CLAUDE.md still earns its keep for Claude-specific rules. If your org relies on Copilot review, re-run review on an active PR and train reviewers to read the new overview sections before arguing about stale threads. Platform owners should map every retiring model id to its replacement in policy and docs now, because default enablement only helps if the global default is still on. Anyone on Bedrock, Vertex, Foundry, or a Claude apps gateway should treat v2.1.278 as the pin and confirm Auto mode server in `/status` before assuming classifier requests are free.

If you only do three things this morning, pin Claude Code to v2.1.278 on one AGENTS.md-only repo, re-request Copilot review after fixing Open findings on one PR, and publish an internal Oct 19 model migration checklist for the six retiring Copilot models.

Tutorial

Pin Claude Code to v2.1.278 and verify AGENTS.md loads when CLAUDE.md is absent. Use a throwaway folder (or a branch) so you do not disturb a production CLAUDE.md. Keep secrets in the environment.

1. Pin `@anthropic-ai/claude-code@2.1.278` and confirm `claude –version`. 2. Create a temp folder with an AGENTS.md that names a test command, and intentionally leave out CLAUDE.md. 3. Run a one-shot prompt that asks Claude to quote the test command and say whether it read AGENTS.md or CLAUDE.md. 4. In interactive mode, open `/config` Project instructions and confirm the AGENTS.md fallback mode. On GitHub, re-request Copilot review after fixing an Open finding, and list which of the six Oct 19 models your org still allows.

npm install -g "@anthropic-ai/claude-code@2.1.278"
claude --version

WORKDIR="$(mktemp -d /tmp/agents-md-probe.XXXXXX)"
cd "$WORKDIR"
printf '%s\n' '# AGENTS.md' '' '- Prefer pnpm over npm.' '- Run tests with: pnpm test' > AGENTS.md

claude -p "Quote the test command from this project's agent instructions file, then say whether you read AGENTS.md or CLAUDE.md."

# Optional gateway opt-out if your proxy cannot pass server-side auto-mode checks:
# export CLAUDE_CODE_AUTO_MODE_SERVER=0

npm install -g "@anthropic-ai/claude-code@2.1.278"
claude --version

WORKDIR="$(mktemp -d /tmp/agents-md-probe.XXXXXX)"
cd "$WORKDIR"
printf '%s\n' '# AGENTS.md' '' '- Prefer pnpm over npm.' '- Run tests with: pnpm test' > AGENTS.md

claude -p "Quote the test command from this project's agent instructions file, then say whether you read AGENTS.md or CLAUDE.md."

# Optional gateway opt-out if your proxy cannot pass server-side auto-mode checks:
# export CLAUDE_CODE_AUTO_MODE_SERVER=0

Recommended AI prompt

Copy this paragraph into ChatGPT, Claude, Gemini, Grok, or whatever you use.

You are my staff engineer for Claude Code instruction-file strategy, Copilot code-review rollout, Copilot model migration, and Claude Code fleet pins on 2026-09-19. Claude Code v2.1.277 reads AGENTS.md when CLAUDE.md is absent (not yet on Bedrock, Vertex, or Foundry). Copilot code review GA groups findings into Open, Resolved since last review, and Previously missed, with smarter auto-resolve and smart batch commit messages. Six Copilot models retire on Oct 19 with named replacements (Gemini 3.8 Flash, GPT-5.6 Sol, GPT-5.6 Luna, Grok 4.6). Claude Code v2.1.278 defaults auto mode to a server-side classifier with no classifier overhead charge when the server performs checks. Ask which Claude plan, gateway topology, Copilot policies, and multi-agent repos we run. Then produce an AGENTS.md versus CLAUDE.md decision checklist, a Copilot re-review checklist using the new overview sections, an Oct 19 model migration table for our enabled models, and a v2.1.278 pin plus auto-mode server verification note. Keep it concrete and copy-paste ready.