Claude Code v2.1.178 Adds Tool(param:value) Permission Rules, Fixes Subagent & Auth Issues

Claude Code v2.1.178 ships with granular permission rules via Tool(param:value) syntax, improved nested .claude/ directory handling, and a hefty list of bug fixes targeting subagents, authentication, and CLI stability.

Permission Rules: Tool(param:value)

You can now match tool input parameters in permission rules using Tool(param:value) syntax with * wildcard support. For example:

Agent(model:opus)

blocks subagents using the Opus model. This closes a gap where subagent spawns in auto mode weren't evaluated by the classifier before launch — they now are.

Nested .claude/ Directories

Skills in nested .claude/skills directories now load when you're working on files in that subtree. On name clashes, the nested skill appears as <dir>:<name> so both remain available. Similarly, agent, workflow, and output-style closest to the working directory win on collision, and project-scope workflow saves target the nearest existing .claude/workflows/.

Workflow Keyword Highlighting

The workflow prompt keyword now uses a purple shimmer highlight and triggers only on explicit phrases like "run a workflow" or "workflow:", not any mention of the word.

/doctor Improvements

The /doctor command now uses a consistent flat tree layout across all sections, clearer section status icons, and highlighted command names.

Bug Fixes

• CLI crash: Fixed out-of-memory crash when inheriting stale websocket/OAuth file-descriptor environment variables from a parent process.
• Chrome silent failure: Fixed Claude in Chrome failing to connect when the OAuth token belongs to a different account than the Claude Code login.
• Subagent fixes: Viewing a subagent's transcript now shows tool results and live progress. Messages sent while it finishes its turn are no longer dropped. Backgrounding a running subagent with Ctrl+B no longer restarts it from scratch.
• Auth cache: Fixed stale cached request configuration causing model requests to continue failing with auth errors after credentials were refreshed outside the session.
• Compaction: Fixed --fallback-model not being honored during compaction — now falls back to the configured fallback model chain on overload or model-availability errors.
• Background sessions: Sessions created with /bg or ←← after a turn finished no longer show "Working" forever in the agents list.
• MCP disallowedTools: Fixed mcp__server, mcp__server__*, and mcp__* specs in subagent disallowedTools being silently ignored.
• Vim mode undo: u now steps through NORMAL/VISUAL-mode commands one at a time instead of merging quick successive commands into a single undo step.
• 401 fix: Fixed claude agents workers failing with 401 Invalid bearer token when the daemon started from a shell with custom ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN.

/bug Enhancement

/bug now requires a description before submitting, and no longer uses model-refusal text as the GitHub issue title.

VSCode-Specific Fix

Pressing Esc to dismiss a CJK IME candidate window no longer cancels the running Claude task.