The Claude Code Changelog

Claude Code v2.1.176 tightens file access controls with corrected path specifiers, making rules like deny, ask, and allow work properly for sensitive files such as .ssh and .env. The hosts also cover handy tmux and SSH quality-of-life fixes, plus why this update is a must for safer terminal workflows.

We dig into undocumented gems in Claude Code v2.1.176, including footerLinksRegexes for clickable terminal badges, plus new quality-of-life tweaks like smoother scrolling controls and localized session titles. We also cover the improved AWS Bedrock credential caching that finally respects real token expiration for longer, less annoying dev sessions.

This episode breaks down Claude Code v2.1.176, including more reliable hook path matching for safer tool access, plus session title language matching that follows your conversation language. It also covers improved AWS Bedrock credential caching, making enterprise workflows smoother by reducing premature session interruptions.

We break down the revamped /usage command in Claude Code v2.1.174, including detailed token attribution across skills, plugins, subagents, and cache misses. Then we cover v2.1.175’s enforceAvailableModels setting for tighter enterprise model control, plus a few handy terminal and exit-time fixes.

We unpack Anthropic’s latest Claude Code update, where sub-agents can recursively spawn deeper agents, backed by a massive one-million-token context window. The conversation also covers token compaction, the new /plugin search bar, AWS Bedrock region fallback, and the risks of letting terminal automation grow into a full recursive agent tree.

We dig into Claude Code v2.1.169’s new background-agent controls, including --json, --all, and the waitingFor field for spotting blocked automations. The discussion also covers enterprise hardening with CLAUDE_CODE_DISABLE_BUNDLED_SKILLS and the trade-offs between tighter security and losing built-in capabilities.

This episode covers the latest Claude Code update, including the new /cd command for switching directories without losing prompt cache and the safe mode option for isolating bugs caused by local plugins, rules, and hooks. It also touches on macOS rendering improvements, a Windows claude -p hang fix, and enterprise support for OpenTelemetry client certificates.

We break down Anthropic’s new managed configuration scopes for Claude Code, including system-level version limits that keep teams on approved builds. Plus, we cover update UX improvements and safer slash-command autocomplete in the terminal.

We break down Claude Code 2.1.163’s new hook behavior, where failed Stop and SubagentStop validations can feed compiler or linter output back into Claude for a self-correcting loop. The episode also covers new plugin list filters and a handy /btw clipboard shortcut for faster terminal workflows.

We break down Anthropic’s latest Claude Code updates, including ordered fallback models for handling rate limits, new controls for disabling thinking tokens, and stronger permission prompts around sensitive system files. Also covered: the handy /btw clipboard command and why these small workflow improvements make agentic terminal work smoother and safer.

This update covers smarter lifecycle hooks that can return structured context directly to the model, making stop-phase workflows cleaner for testing and automation. It also fixes temporary directory handling for local builds and improves session ID propagation for stdio MCP servers when using resume and continue flows.

This episode covers the latest v2.1.162 update, where WebFetch now honors explicit allowlists and blocklists over built-in domain rules for stronger control. It also highlights safer slash command autocomplete, read-only config fallbacks for containers, and the new waitingFor JSON key for better agent automation.

This episode breaks down Claude Code v2.1.161’s new fault-tolerant parallel execution, where one failed tool call no longer kills the whole batch. The hosts also cover OpenTelemetry resource tagging, secret redaction in MCP listings, and improved terminal rendering in VS Code with the new /terminal-setup command.

We break down the new permission safeguards in Claude Code v2.1.160, including explicit consent for writes to critical shell and config files like .npmrc and .zshenv. The episode also covers the faster read-before-edit check, token-saving grep behavior, and the rename of the dynamic workflow trigger to ultracode.

We break down Claude Code’s new Dynamic Workflows system, where a local JavaScript orchestrator replaces context-heavy multi-agent prompting for faster, cleaner execution. The episode also covers 16-agent parallel runs, the 1,000-invocation safety cap, Opus 4.8’s high-effort default, and the deprecation of the old fast-mode override in favor of /model and /fast commands.

We break down Claude Code 2.1.158’s new Auto Mode support for Bedrock, Vertex AI, and Palantir Foundry, and what it means for developers working inside tightly controlled enterprise environments. Plus: the new OpenTelemetry audit flag, full autonomous write-run-correct loops, and the fix for stubborn Git worktree locks.

We break down Claude Code v2.1.157’s local .claude/skills workflow, where executable scripts can be loaded straight from a developer’s machine for instant tool creation and hot reloading. The conversation also digs into the security risks of committing custom plugins to source control, including the possibility of arbitrary code execution.

We dig into Claude Code’s new /code-review --fix workflow, which can edit files directly instead of just leaving comments, and the revamped /simplify pass for trimming dead code and redundant structure. The episode also covers safer usage tips, from keeping a clean git tree to reviewing diffs, plus the new skipLfs option for speeding up large repos.

We break down a critical credential leak in custom API gateways, where Claude Code mistakenly sent primary Anthropic OAuth credentials to the wrong place. The episode also covers faster git and GitHub plugin loading with skipLfs, plus new safeguards for subagent MCP configs to prevent privilege escalation.

This episode digs into the v2.1.132 Claude Code update, including a stdio MCP memory leak that could balloon long sessions past 10GB, plus the practical differences between stdio and HTTP transport. It also covers new session-aware Bash env vars, a terminal escape hatch for alternate screen mode, and a permission-mode bug that could quietly override your intent when resuming plan-mode sessions.

This episode breaks down how Claude Code v2.1.129 changes plugin loading with URL-based zip installs, moving plugins from local setup to a true distribution model. It also explores the new experimental themes and monitors fields, and the security tradeoffs that come with treating plugins like packaged software.

This episode breaks down a subtle Git behavior in Claude Code v2.1.128 where new worktrees now branch from local HEAD instead of origin/default-branch, preventing unpushed commits from being skipped. It also covers related reliability improvements, including OTEL isolation for subprocesses and better handling of parallel tool calls.

This episode explores how Claude Code’s `ultrareview` command uses exit codes, streamed output, and PR targeting to make AI review behave like a real CI check. It also digs into the multi-agent verification model, cost tradeoffs, and the network constraints teams need to consider before wiring it into GitHub Actions.

We break down Anthropic’s new xhigh effort level for Claude Code on Opus 4.7, why it now defaults for Max subscribers, and how the new slider and config flow make it easier to use. The discussion also covers when deeper reasoning is worth the extra latency and cost, plus how to spot the tasks where xhigh really pays off.

Claude Code now asks your configured ANTHROPIC_BASE_URL for /v1/models, so the model picker reflects the IDs your gateway actually exposes instead of a hardcoded Anthropic-only list. The episode breaks down why that matters for LiteLLM, Bedrock, and other proxies, plus the catch: the backend still has to return truthful model names or the client will pass them through verbatim.