Claude Code Changelog

Claude Code CLI 2.1.126 changelog: Fixes: • Fixed pasting an image larger than 2000px breaking the session — images are now downscaled on paste, and oversized images in history are automatically removed and the request retried • Fixed showing the login screen for "OAuth not allowed for organization" errors — now shows guidance to contact your admin • Fixed OAuth login failing with timeout on slow or proxied connections, in IPv6-only devcontainers, and when the browser callback can't reach localhost • Fixed a rare race where a concurrent credential write could clear a valid OAuth refresh token • Fixed API retry countdown sticking at "0s" instead of counting down between attempts • Fixed "Stream idle timeout" error after waking Mac from sleep mid-request • Fixed background and remote sessions falsely aborting with "Stream idle timeout" during long model thinking pauses • Fixed a hang where the assistant could finish thinking but show no output after a run of empty turns • Fixed overly fast trackpad scrolling in Cursor and VS Code 1.92–1.104 integrated terminals • Fixed claude.ai MCP connectors being suppressed by manual servers stuck in needs-auth state • Fixed Japanese/Korean/Chinese text rendering as garbled characters on Windows in no-flicker mode • Fixed Ctrl+L clearing the prompt input — it now only forces a screen redraw, matching readline behavior • Fixed deferred tools (WebSearch, WebFetch, etc.) not being available to skills with context: fork and other subagents on their first turn • Fixed plan-mode tools being unavailable in interactive sessions launched with --channels • Fixed blank remote-session transcript when certain messaging tools are unavailable • Fixed /plugin Uninstall reporting "Enabled" instead of "Uninstalled" • Fixed /remote-control retries appearing stuck on "connecting…" — each retry now shows its result, and unenrolled trusted-device failures are caught up-front • Fixed Remote Control failure notification not showing the error reason for initial connection failures • Fixed Agent SDK hang when the model emits a malformed tool name in a parallel tool call batch Security/safety changes: • --dangerously-skip-permissions now bypasses prompts for writes to .claude/, .git/, .vscode/, shell config files, and other previously-protected paths (catastrophic removal commands still prompt as a safety net) • claude auth login now accepts the OAuth code pasted into the terminal when the browser callback can't reach localhost (WSL2, SSH, containers) • Auto mode: the spinner now turns red when a permission check stalls, instead of looking like the tool is running • Security: Fixed allowManagedDomainsOnly / allowManagedReadPathsOnly being ignored when a higher-priority managed-settings source lacked a sandbox block Other changes: • The /model picker now lists models from your gateway's /v1/models endpoint when ANTHROPIC_BASE_URL points at an Anthropic-compatible gateway • - Added claude project purge [path] to delete all Claude Code state for a project (transcripts, tasks, file history, config entry) — supports --dry-run, -y/--yes, -i/--interactive, and --all • claude_code.skill_activated OpenTelemetry event now fires for user-typed slash commands and carries a new invocation_trigger attribute ("user-slash", "claude-proactive", or "nested-skill") • Host-managed deployments (CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST) no longer auto-disable analytics on Bedrock/Vertex/Foundry • Windows: PowerShell 7 installed via the Microsoft Store, MSI without PATH, or .NET global tool is now detected • Windows: when the PowerShell tool is enabled, Claude now treats PowerShell as the primary shell instead of defaulting to Bash • Read tool: removed the per-file malware-assessment reminder that could cause spurious refusals and "this is not malware" commentary on legacy models • Bounded total size of file-modified reminders when a linter touches many files at once • Windows: clipboard writes no longer expose copied content in process command-line arguments visible to EDR/SIEM telemetry; also fixes >22KB selections not reaching the clipboard • PowerShell tool: bare -- (e.g. git diff -- file) is no longer mis-flagged as the --% stop-parsing token Source: #21126" target="_blank" rel="nofollow noopener">github.com/anthropics/cla…

Claude Code 2.1.126 additional updates CLI surface: Added: • models: claude-empty- Removed: • models: claude-empty-I File: #diff-662031a066e433468319e799350331e143e4635468b9c2924019d16654027e31L5-R5" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Metadata: • Time since 2.1.124 release: 15h 35m 24s • Bundle file size: -1.7 kB (-0.0%) • Prompt files: +1 (+1.3%) • Prompt tokens: +5,142 (+6.0%) Prompt token mix by kind: • system-reminder: 35.1%→32.6% • system: 33.3%→31.4% • tools: 20.4%→25.4% • agent: 5.8%→5.5% • skill: 3.0%→2.8% • system-data: 2.4%→2.3% Files: • metadata: #diff-a08dfa163baef97ea4f56d29009ac65f5f7dbc46e3a58e80d7aa06a2a2520899L10-R10" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… • prompt stats: #diff-fa8e31336ece71d159f37b6fd4ed0c61939849140bd5c08c7ca793b5b2ae7f58L1-R1" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Other prompt/string highlights: 1) Edit tool performs exact string replacements in files. Links: 1) #diff-d98c9d5750be0d3d005e4bf2b7153b7a85d658c33884e9c9ef520cd58aa2536cR3" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… More details: github.com/marckrenn/clau…

Claude Code 2.1.126 has been released. 33 CLI changes Highlights: • Edit tool now performs exact string replacements in files, so edits only change exact matched text • claude project purge [path]: deletes Claude Code data—transcripts/tasks/files/history/config; --dry-run/-i • --dangerously-skip-permissions disables prompts for .claude/.git/.vscode and shell RCs, allowing writes Full details are in thread ↓

Claude Code 2.1.126 is about to be released #cccnext

Claude Code 2.1.124 is about to be released #cccnext

Claude Code 2.1.123 additional modifications CLI surface: Added: • models: claude-empty-U Removed: • models: claude-empty- File: #diff-662031a066e433468319e799350331e143e4635468b9c2924019d16654027e31L917-R917" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Metadata: • Time since 2.1.122 release: 23h 3m 21s • Bundle file size: -0.2 kB (-0.0%) • Prompt files: +3 (+4.1%) • Prompt tokens: +6,659 (+9.4%) Prompt token mix by kind: • system: 40.1%→36.6% • system-reminder: 26.7%→31.0% • tools: 19.8%→20.0% • agent: 7.0%→6.4% • skill: 3.6%→3.3% • system-data: 2.9%→2.7% Files: • metadata: #diff-a08dfa163baef97ea4f56d29009ac65f5f7dbc46e3a58e80d7aa06a2a2520899L10-R10" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… • prompt stats: #diff-fa8e31336ece71d159f37b6fd4ed0c61939849140bd5c08c7ca793b5b2ae7f58L1-R1" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Other prompt/string highlights: 1) Selects and attaches relevant memory files for user queries. Links: 1) #diff-9b0379aa998142e7e9a5916702866b91652bf360465c272153861c476541431dR22" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… More details: github.com/marckrenn/clau…

Claude Code 2.1.123 system prompt updates Notable changes: 1) Claude no longer has the upfront rule limiting security help to clearly authorized contexts (pentests/CTFs/defensive) and explicitly refusing destructive techniques, DoS, mass targeting, supply-chain compromise, or malicious evasion. This loosens safety gating for security requests. 2) Claude no longer has the explicit constraint to “NEVER generate or guess URLs” except when confidently programming-related (and otherwise stick to user-provided URLs). Expect more willingness to invent/share links without that hard prohibition. 3) Claude gains a new “System” block: user-facing text is rendered as GitHub-flavored Markdown (CommonMark, monospace); tools run under user-selected permissions and Claude must not repeat an identical denied tool call; system tags may appear; suspected tool-output prompt injection must be flagged; prior messages may be auto-compressed near context limits. Links: 1) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L63" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… 2) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L64" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… 3) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1R61" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau…

Claude Code 2.1.123 has been released. 1 CLI change, 3 system prompt changes Highlights: • Fixed OAuth 401 retry loop when CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 is set, restoring CLI authentication • Responses may now include generated or guessed URLs, which can point to incorrect external sites Full details are in thread ↓

Claude Code 2.1.123 is about to be released #cccnext

Claude Code 2.1.122 additional updates CLI surface: Added: • env vars: ANTHROPIC_BEDROCK_SERVICE_TIER, CLAUDE_CODE_INVESTIGATE_FIRST, CLAUDE_COWORK_MEMORY_INDEX_CONTENT, DISABLE_BRIEF_MODE_STOP_HOOK Removed: • env vars: CLAUDE_COWORK_MEMORY_SKIP_DEFAULT_ENTRYPOINT File: #diff-662031a066e433468319e799350331e143e4635468b9c2924019d16654027e31L7-R7" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Metadata: • Time since 2.1.121 release: 23h 59m 4s • Bundle file size: +90.8 kB (+0.2%) • Prompt files: +3 (+4.3%) • Prompt tokens: +1,494 (+2.1%) Prompt token mix by kind: • system: 40.4%→40.1% • system-reminder: 27.2%→26.7% • tools: 18.6%→19.8% • agent: 7.1%→7.0% • skill: 3.6%→3.6% • system-data: 3.0%→2.9% Files: • metadata: #diff-a08dfa163baef97ea4f56d29009ac65f5f7dbc46e3a58e80d7aa06a2a2520899L10-R10" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… • prompt stats: #diff-fa8e31336ece71d159f37b6fd4ed0c61939849140bd5c08c7ca793b5b2ae7f58L1-R1" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Other prompt/string highlights: 1) Adds a system reminder that restricts the assistant to malware-analysis-only guidelines when examining suspected malware. Links: 1) #diff-05489bf6351e53a9f4435615f00ebcddd2fd9df60ed3a1b7683a7565703b7f0bR13" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… More details: github.com/marckrenn/clau…

Claude Code 2.1.122 system prompt updates Notable changes: 1) Claude no longer receives the early “System/Doing tasks” instructions: the explicit tool list (Read/Edit/Write/Glob/Grep/Bash/TodoWrite), the ban on git interactive flags, guidance to use gh for GitHub ops, the {rerun:'bN'} mechanism, and Bash persistence/timeout details were removed. 2) Claude gains a new safety preface: “Read/search/investigate freely—looking is not acting,” but confirm hard-to-reverse or shared-system actions unless durably authorized, and prior approval doesn’t carry over. The line also contains stray ";return#" text, suggesting prompt corruption. Links: 1) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L66" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… 2) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L80-R66" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau…

Claude Code 2.1.122 has been released. 18 CLI changes, 2 system prompt changes Highlights: • Responses prepend 'looking is not acting' and require explicit consent before risky actions, preventing harm • Added ANTHROPIC_BEDROCK_SERVICE_TIER to select Bedrock tier (default/flex/priority), affects tier, speed, cost • Pasting a PR URL into /resume maps to its creating session (GitHub/GHE/GitLab/Bitbucket), easing lookup Complete details in thread ↓

Claude Code 2.1.122 is about to be released #cccnext

Claude Code 2.1.121 additional updates CLI surface: Added: • commands: prune • options: --prune, --yes, -y • env vars: CLAUDE_CODE_DISABLE_AGENTS_FLEET, CLAUDE_COWORK_MEMORY_SKIP_DEFAULT_ENTRYPOINT Removed: • env vars: CLAUDE_CODE_LEAN_PROMPT, INVOCATION_ID File: #diff-662031a066e433468319e799350331e143e4635468b9c2924019d16654027e31L5-R5" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… Metadata: • Time since 2.1.120 release: 2d 6h 31m 38s • Bundle file size: +272.6 kB (+0.7%) • Prompt files: +1 (+1.4%) • Prompt tokens: +443 (+0.6%) Prompt token mix by kind: • system: 40.2%→40.4% • system-reminder: 27.2%→27.2% • tools: 18.7%→18.6% • agent: 7.2%→7.1% • skill: 3.7%→3.6% • system-data: 2.9%→3.0% Files: • metadata: #diff-a08dfa163baef97ea4f56d29009ac65f5f7dbc46e3a58e80d7aa06a2a2520899L10-R10" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… • prompt stats: #diff-fa8e31336ece71d159f37b6fd4ed0c61939849140bd5c08c7ca793b5b2ae7f58L1-R1" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… More details: github.com/marckrenn/clau…

Claude Code 2.1.121 system prompt updates Notable changes: 1) Claude loses explicit System guidance on tool permission prompts/denials, system-tag handling, and flagging prompt-injection in tool output. In its place, the prompt adds Git workflow constraints: no interactive git flags (-i) and GitHub ops should use the gh CLI. 2) Claude is no longer constrained by a long 'Doing tasks' rubric (no extra features/abstractions, minimal error handling, default no comments, UI must be tried in browser, prefer edits over new files, OWASP vigilance, and verified-vs-assumed reporting). This materially loosens behavior expectations. 3) Claude gains new Bash-operational rules: prior commands should be rerun via a {rerun:'bN'} footer token (not retyped), the working directory persists across calls but shell state (env/functions) does not, and Bash timeouts are defined in milliseconds. Links: 1) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L66-R66" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… 2) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L73-R74" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau… 3) #diff-b0a16d13c25d701124251a8943c92de0ff67deacae73de1e83107722f5e5d7f1L73-R74" target="_blank" rel="nofollow noopener">github.com/marckrenn/clau…

Claude Code 2.1.121 is now available. 39 CLI changes, 3 system prompt changes Highlights: • Added MCP server option alwaysLoad: when true, that server's tools skip search deferral and load immediately • Bash tool drops shell state between runs and adds rerun-footer tokens, so commands don't share prior context • PostToolUse hooks can replace output for all tools, letting hooks override tool results (was MCP-only) Complete details in thread ↓

Claude Code 2.1.121 is about to be released #cccnext