claude-agent-acp pin 0.73.0 bundles Claude Code 2.1.257; claude-opus-5-5 requires 2.1.280 (0.81.0) #1137

Status 2026-10-26

• PR Bump claude-agent-acp pin to 0.81.0 for claude-opus-5-5 #1184 (one commit, 0.73.0 -> 0.81.0) fixes this on main; FrontierPhysics has run claude-opus-5-5 on 0.81.0 since 2026-09-23 (review posted on the PR). PR fix(agents): bump ACP shims so claude-opus-5-5 and gpt-6 models can run #1139 is the same bump — benchflow
• release/0.8 fixed it differently (6bd05f2: adapter 0.81.2 plus Claude Code 2.1.280 through CLAUDE_CODE_EXECUTABLE); not in a PyPI release (Ship the release/0.8 fixes in a Python release, or cherry-pick them to main #1197).
• FrontierPhysics-Infra re-pins through FPINFRA_CLAUDE_AGENT_ACP_VERSION until a release carries it — Bo

bench eval run --agent claude-agent-acp --model claude-opus-5-5 fails on every turn with:

ACP error -32603: Internal error: API Error: 400 Claude Code 2.1.257 does not support this model; version 2.1.280 or newer is required. Run 'claude update', or update the Claude desktop app, then try again.

benchflow 0.7.8 pins @agentclientprotocol/claude-agent-acp@0.73.0 (benchflow/agents/registry.py, _CLAUDE_AGENT_ACP_PACKAGE), which bundles Claude Code 2.1.257 through @anthropic-ai/claude-agent-sdk@0.3.257. claude-agent-acp@0.81.0 bundles 2.1.280 (claude-agent-sdk@0.3.280) and runs claude-opus-5-5 (verified 2026-09-23, Daytona sandbox, agent/runtime.json sandbox.claude_code_version = 2.1.280).

• raise _CLAUDE_AGENT_ACP_PACKAGE to @agentclientprotocol/claude-agent-acp@0.81.0 (or make the pin overridable, e.g. BENCHFLOW_CLAUDE_AGENT_ACP_VERSION) — benchflow
• until then FrontierPhysics-Infra re-pins it per run through FPINFRA_CLAUDE_AGENT_ACP_VERSION (mirror of FPINFRA_CODEX_ACP_VERSION) — Bo