Claude Code v2.1.295 Adds onFailure Blocking Hooks and Program Status Protocol
Claude Code v2.1.295 landed with a mix of new hook controls, gateway tuning options, and a batch of MCP and headless-mode fixes. The most notable addition is onFailure: "block" for command and HTTP hooks — if a hook can't start, times out, or exits with an unexpected code, the action is blocked instead of being let through. That inverts the previous behavior, where a broken hook was effectively a no-op.
Terminal and CLI changes
- Program Status Protocol (OSC 7501) support: terminals implementing it can display whether Claude Code is working, waiting on you, or done.
/copynow includes quoted text in the picker, so a drafted message copies without its>markers.claude -pprints a line on stderr (when stderr is a terminal) explaining what the run is waiting for if it stays open after its last turn.- A "Backgrounding cancelled" message when you stop the turn while
←is waiting on the current tool. claude plugin install,enable,disable, andmarketplace addwarn when the settings file they write to doesn't load.
Gateway and upstream controls
Gateway upstreams (Bedrock, Vertex, Foundry, and other cloud backends) pick up two new settings:
timeouts.upstream_ttfb_ms models: ["model-id", "other-*"]
timeouts.upstream_ttfb_ms limits how long a stream may take to start before failing over or returning a 502. The optional models list restricts which models get sent to a given upstream — including on failover — and a single * in an entry acts as a wildcard. Successful inference responses now carry a request-id header so request_id in Claude Code telemetry matches the gateway audit log, and inference audit events gained upstream_request_id from Bedrock, the Anthropic API, or another upstream.
For machines without managed settings, user settings now support forceLoginMethod: "gateway" and forceLoginGatewayUrl, so /login opens against that gateway.
Notable fixes
- Requests on
[1m]models no longer fail wholesale when a gateway, Bedrock, Vertex, or Foundry refuses the context-1m beta — Claude Code resends without it. claude -ptext output no longer drops earlier responses when background work starts another turn; each turn's response prints when the turn ends.- Remote MCP servers in headless and SDK sessions reconnect properly after outages longer than 15 seconds. Servers that drop every connection right after it connects now back off up to 30s instead of tight-looping.
- MCP servers repeating a pagination cursor are no longer asked for the same page up to 20 times per connect.
- CSS, JavaScript, and XML files returned by MCP tools no longer save as
.bin(which the Read tool refuses). - Bash tool calls work when the model passes
command_descriptioninstead ofdescription.
For mod authors, $.ui.notify raises a native notification through your notification setting and reports the channel; a mod's Button now accepts children (strings and Text) so a list row is one pressable with a chip or dim detail inside. CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS caps how long unattended retry mode waits out 429 and 529 errors.
📖 Read the full source: GitHub Claude-Code
👀 See Also
Claude Code v2.1.252 Fixes Bash Errors, Remote Control Stalls
Claude Code v2.1.252 fixes Bash failures on Macs, 'always allow' persistence, Remote Control stalls in Claude Desktop/VS Code, and oversized background notifications.
OpenClaw Updates Break Setups: What Users Are Doing About It
A user managing 5 sites with OpenClaw, Claude Code, and Codex shares that updates often break their working setup, prompting a cautious approach to upgrading.
Claude Boosts Riemann Hypothesis Zero Bound from 41.6% to 67.2%
An unreleased research version of Claude improved the lower bound for zeros on the critical line from 41.6% to 67.2%, using a novel combination of prior work.

Reddit Discussion on Claude's Impact on MVP Development and Founder Pitfalls
A Reddit user discusses how Claude AI lowers technical barriers for building MVPs from $3k-$5k to DIY, but warns about increased competition and founders focusing too much on building versus marketing, PMF, and operations.