What Claude Code's May 1 Update Means for Webflow Partners

Anthropic shipped a broad Claude Code update on May 1, 2026. The release notes run long and read like a maintenance changelog at first glance. Most of the items are quality-of-life fixes. A few are quietly significant for solo Webflow Partners and small studios who use Claude Code as part of their daily workflow. This piece sorts the noise from the signal, calls out the items worth adopting this week, and flags the ones safe to ignore until a future release. The discipline is to adopt selectively rather than reflexively.

What Did Anthropic Ship in the May 1 Claude Code Release?

The release was broad rather than headline-grabbing. Smarter model picking with the slash-model command now lists models from a configured Anthropic-compatible gateway. A new claude project purge command deletes Claude Code state for a project, including transcripts, tasks, file history, and config entries. Stronger permission handling, improved OAuth login, Windows and PowerShell fixes, better OpenTelemetry telemetry, and stability work across sessions, retries, and terminal behavior round out the notes.

The same release window also retired the 1M token context beta on Claude Sonnet 4.5 and Claude Sonnet 4. The beta header context-1m-2025-08-07 now has no effect on those models, and requests over the standard 200,000-token context window return an error. To use the 1M context window, the migration path is to Claude Sonnet 4.6 or Claude Opus 4.6, where it is generally available at standard pricing.

What Should Solo Webflow Partners Adopt First?

Three changes earn immediate attention. The claude project purge command, because it gives Partners a clean way to delete project state when an engagement ends or a contractor rotates off. The OpenTelemetry improvements, because numeric attributes now emit as numbers rather than strings, which makes audit logs queryable in any modern observability tool. The PR-URL search in slash-resume, because it lets Partners jump back into the session that created a specific pull request, which is the moment-of-most-friction in async code review.

Each of these has a measurable impact on a solo practice within the first week of adoption. The project purge alone solves a problem most Partners did not realize they had. Stale Claude Code state from old engagements clutters the local environment and creates confusion when a new project shares a directory name with a closed one. Spending fifteen minutes purging four old projects on Monday morning is the cheapest workflow upgrade in the entire release. I covered the upstream discipline in my Claude Code skills piece.

What Does the Project Purge Command Actually Do?

The command claude project purge takes a path argument and deletes all Claude Code state associated with that project. The state includes session transcripts, task records, file history, and the project's entry in the config. The release added flags for safe operation, including dry-run, yes, interactive, and an all flag for batch cleanup. The dry-run flag prints what would be deleted without actually deleting anything, which is the safe way to preview the operation before committing.

For a typical Webflow Partner, the cleanup pattern that works is to run the purge with dry-run on each closed engagement at the end of every quarter. Confirm the deletion looks right. Then run again without dry-run. The whole operation takes ten minutes per quarter and keeps the local Claude Code environment focused on active work. Partners who skip this accumulate state from twenty closed engagements over a year, which slows search, clutters resume options, and occasionally surfaces stale context in active sessions.

What Changed With Model Picking, and Does It Matter for a Solo Practice?

The slash-model picker now lists models from a gateway's v1/models endpoint when the ANTHROPIC_BASE_URL environment variable points at an Anthropic-compatible gateway. For Partners running Claude Code through a corporate gateway, an LLM cost-tracking proxy, or a self-hosted observability layer, this is a significant quality-of-life improvement. The picker reflects what is actually available rather than what the local Claude Code binary thinks should be available.

For solo Partners using direct Anthropic API access without a gateway, the change is invisible. The default behavior matches what shipped before. The signal worth noting is that Anthropic is treating gateway-based access as a first-class deployment mode, which suggests gateway tooling will get more first-class support in future releases. Partners who eventually want to centralize their AI cost across multiple clients should plan for a gateway adoption in the next six to twelve months. I covered the cost lens in my monthly AI tooling cost piece.

What Are the OpenTelemetry Changes, and Why Should I Care?

OpenTelemetry is the open standard for observability data. Claude Code emits structured logs in OpenTelemetry format, which means Datadog, Honeycomb, Grafana, and most other observability tools can ingest those logs natively. The May 1 release fixed two specific issues. Numeric attributes on api_request and api_error log events now emit as numbers rather than strings, which makes them filterable and aggregatable in queries. A new claude_code.at_mention log event captures when the agent resolves an at-mention to a file or context.

The combined effect is that Claude Code usage is now properly observable at the per-action level. Partners running structured logging can now query "how many api_error events fired against the Acme project this month" or "which files were at-mentioned most often during the Foo migration." That level of detail did not exist cleanly before May 1. For studios building toward AI audit-log discipline, the change is foundational. I covered the audit-log architecture in my AI audit logs piece from earlier today.

What About the 1M Context Retirement on Sonnet 4.5 and 4?

The retirement is a deprecation event, not a capability loss. Anthropic kept the 1M context window. It just moved the GA version to Sonnet 4.6 and Opus 4.6. Partners using Sonnet 4.5 or Sonnet 4 with the beta context-1m header for long-context tasks need to migrate to 4.6 to keep that capability. The pricing remains standard, with no beta header required. The migration is essentially a model name change in configuration plus a regression test.

The strategic signal is that Anthropic is consolidating support around the 4.6 family, which makes Sonnet 4.6 the default safe choice for new client engagements and Opus 4.7 the default for high-stakes work. Partners standardizing on Sonnet 4.6 today will get a stable foundation that ages well through the rest of 2026. Partners still on older versions for specific reasons should plan a migration window before more deprecations land.

What Should I Skip From This Release?

Three categories of changes are safe to ignore for a typical solo Webflow practice. The Bedrock service tier environment variable, ANTHROPIC_BEDROCK_SERVICE_TIER, only matters for Partners using AWS Bedrock as the model backend, which is a small minority. The Windows and PowerShell fixes only matter on Windows, which most Webflow Partners do not run as their primary development environment. The bulk of the stability fixes are background polish that takes effect automatically without configuration.

The cleanup work and forks-from-rewound-timelines fix are technical wins that the typical Partner will simply benefit from without thinking about. Reading the full changelog to understand each fix is rarely the right use of time. The pattern that works is to scan the headlines, identify the two or three items that fit the studio's current workflow, adopt those, and skip the rest until a future release surfaces them again. I covered the rhythm of selective adoption in my AI as senior team member framework piece.

How Does This Stack With Cursor's May 1 Update?

Cursor and Claude Code both shipped meaningful updates on May 1, 2026. Cursor added first-party plugin marketplace controls with three distribution modes. Claude Code added smarter model picking, project purge, and OpenTelemetry improvements. The two are not competing for the same workflow. Cursor's plugin marketplace is about distributing tools across a team. Claude Code's release is about depth and observability of individual sessions.

For Partners using both, the practical effect is that the toolchain matured noticeably in a single day. The plugin distribution layer is now manageable across contractors. The session-level observability is now production-grade. The combination supports the kind of audit-trail discipline that procurement teams ask about. The synchronicity is unlikely to be coincidence. The vendor convergence on governance-grade tooling reflects what enterprise customers are demanding from agent-augmented workflows. I covered the parallel piece in my Cursor plugin marketplace piece.

What Is the Single Action Item Worth Doing This Week?

Run claude project purge with dry-run against three of your closed engagements. Read the dry-run output. If it looks right, run the purge for real. The whole exercise takes thirty minutes and produces immediate clarity in your local Claude Code environment. The benefit compounds over the rest of the year because every new engagement starts in a cleaner state.

The deeper benefit is that the act of running the purge forces you to think explicitly about which projects are closed and which are still active. Most solo Partners carry mental ambiguity about engagement boundaries that surfaces as procrastination on cleanup work. The purge command forces the closure to be explicit. That alone is worth the thirty minutes, even ignoring the file-system benefits. The Anthropic news index covered Claude Opus 4.7's launch on April 17 with reported 13 percent improvement on a 93-task internal coding benchmark over Opus 4.6, which is the model most Partners should pair with the purge discipline going forward.

What Should I Watch For in the Next Claude Code Release?

Two trajectories suggest the most likely next moves. Project purge will probably extend with archive-and-restore semantics rather than just delete, because most studios want to keep an audit trail of closed projects rather than wipe them entirely. OpenTelemetry support will probably extend to cover more event types, particularly around tool-use and MCP server interactions, because that is where governance-relevant telemetry is most valuable.

The broader pattern is that Anthropic is converging Claude Code toward enterprise-grade auditability. The May 1 release is one step in that direction. Studios that build governance discipline now will find each subsequent release easier to adopt because the foundation is already in place. Studios that wait will face a steeper learning curve when the discipline becomes a procurement requirement. Anthropic announced its enterprise services joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman on May 4, which signals where the company believes the next year of AI revenue lives. The May 1 release is the engineering side of that same bet.

If you are running a Webflow practice and want to talk through which Claude Code features fit your engagement structure, drop me a line and tell me how many active client repositories you currently maintain in the tool. Let's chat.