Claude Code v2.1.269: plugin eval reports and Bash command diffs#
Published by Alex Sorokoletov
Part of the Claude Code Version Tracker series. | Official Env Vars | Official Changelog
Coming from v2.1.266, a v2.1.269 user can run claude plugin eval to score a plugin's own eval suite against Claude Code and get a reproducible JSON and HTML report, see a diff of the files a Bash command changed folded into the Bash tool result, and raise the Workflow tool's per-run concurrent-agent ceiling for large fan-outs.[1]
What a user gets first#
claude plugin eval runs a plugin's eval suite and returns scored, reproducible results as both JSON and an HTML report, so a plugin author can check behavior before shipping.[1] /output-style [name] lists and switches output styles, and it now works over Remote Control and in headless sessions rather than only at an interactive prompt.[1]
The Bash tool result now includes a diff of the files a command edited, controlled by the bashEditDiffEnabled setting.[1] A CLAUDE_CODE_BASH_EDIT_DIFF override is present in the build.
A run of fixes restores prompt-cache reuse that broke on resume after an interrupt and on the turn after an output-token cutoff.[1] Earlier in the window, maxEffortLevel (v2.1.267) caps the effort level on every provider including Bedrock, Vertex and Foundry.[1]
New environment variables#
| Variable | What It Does |
|---|---|
CLAUDE_CODE_BASH_EDIT_DIFF | Overrides the Bash edit-diff behavior, which the bashEditDiffEnabled setting otherwise controls. Present in v2.1.269; the changelog names the bashEditDiffEnabled setting, not this env var. |
CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS | Raises the Workflow tool's per-run concurrent-agent limit; changelog range 1 to 256. Present in v2.1.269. |
CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS | Extends the LLM gateway /v1/models discovery timeout; default 3 seconds. Present in v2.1.269. |
CLAUDE_CODE_WEBFETCH_DEADLINE_MS | Overrides WebFetch's 300-second request deadline (added v2.1.268). |
What these tell us#
The confirmed surface points at two directions. One is scale and control for agent runs: a Workflow concurrency knob, a plugin eval command that scores a suite into a reproducible report, and an effort ceiling that holds across Bedrock, Vertex and Foundry. The other is prompt-cache correctness: fixes that restore cache reuse on resume after an interrupt and on the turn after an output-token cutoff.
Both directions read as reach for operators and agent runs rather than model reach. The plugin eval command gives an author a repeatable score before shipping, the Workflow limit lets a large fan-out run at a chosen width, and the gateway timeout and WebFetch deadline harden the paths that talk to slow or self-hosted endpoints.
This analysis is conducted for educational and research purposes under fair use principles. All trademarks and software referenced belong to their respective owners.
Sources#
- Claude Code Official Changelog, v2.1.267 through v2.1.269 release notes
Related Versions#
- Claude Code v2.1.247: /claude-api cost tooling and org spinner tips loaded from a file. Admin API cost coverage, file-based spinner tips.
- Claude Code v2.1.245: A Linux Startup Crash Fix and Prompt-Cache TTL Controls. glibc crash fix, cache TTL knobs.
- Claude Code v2.1.241: Artifact Deletion, Lower-Priority Mode, and a Bedrock Proxy Double-Billing Fix. Artifact lifecycle, priority mode.
- Claude Code v2.1.238: Plugin Marketplaces Can Mint Their Own Auth Headers. Marketplace auth headers.
- Claude Code v2.1.237: A Concise Output Style, a Persistent Default Model, and a Silent Turn Reminder. Output styles, persistent model.
Previous Claude Code analysis: v2.1.266 (September 9, 2026)
Next Claude Code analysis: v2.1.283 (September 26, 2026)
All 67 Claude Code analyses: Claude Code Version Tracker · Official Claude Code changelog