Claude Code v2.1.292 Adds Effort Control for Subagents and 20+ Bug Fixes

Claude Code v2.1.292 is out. Most of the diff is bug fixes, but there are a few new knobs worth knowing about.

New flags and parameters

• --marketplace <source> added to claude plugin install. It adds the marketplace if needed — under the same policy checks as claude plugin marketplace add — then installs the plugin from it. One command instead of two.
• The Agent tool now takes an effort parameter, so you can spawn a sub-agent at a specific effort level.
• CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS env var sets a longer base delay for backoff when retrying an overloaded (529) request.

Mod hook additions

• prompt.autocomplete — an event mods can hook to add their own rows to the prompt box autocomplete list.
• Prompt caching for $.model.complete: prompt and system take blocks of text, and cache: true on a block caches the request up to it.
• Workflow agents exposed to the agent.spawn hook with their run and index, so a mod can refuse them.

Security fixes

• PreToolUse hook approvals and auto mode no longer bypass the permission prompt for file reads from network (UNC) paths.
• Sandboxed commands can no longer read staged file copies of /ultrareview uploads under ~/.claude/seed-admin.
• A managed sandbox read-deny path (and adjacent user ones) appearing or re-pointing mid-session now correctly drops project grants inside it and ends credential injection from covered files.
• Notebook/PDF reads on macOS and Windows can no longer return a file outside what was approved via a link swapped in mid-read.
• A tampered on-disk cache of server-managed settings can no longer switch off or unseat the built-in policy plugin while the settings fetch fails.
• rm -rf on the 8.3 short name or another alternate Windows spelling of the home folder or a drive is now treated as removing it.

Notable behavior fixes

• Subagent definitions with permissionMode: auto no longer enter auto mode when it's unavailable (disabled by settings, circuit breaker, or an unsupported model).
• NO_PROXY is no longer ignored for Claude Code's own API requests (sign-in, policy, feedback, artifacts) when HTTPS_PROXY is set.
• An MCP tool name over 128 characters no longer fails every request; that tool is left out and an MCP error names it.
• claude plugin commands like marketplace add and install no longer run before an org's managed settings load on first run.
• One-shot claude -p and Agent SDK runs now wait for background commands and scheduled wakeups instead of killing them 5 seconds after the final result.
• Plan mode is restored correctly when resuming from the claude --resume picker or /resume.
• Scheduled tasks created after /resume, /branch, or /clear now fire.
• A background session's /loop no longer silently stops after a process restart.
• Grep and Glob retry once or tell you when the given file/folder can't be read, instead of reporting no matches.
• The Read tool errors when a PDF's pages is a list like "6,9,15", instead of silently returning only the first entry.
• @-mentioned text files over 256KB now tell Claude the file's size and to read it in portions.

If you're running agents in sandboxes, the UNC-path and sandbox read-deny fixes are the ones to prioritize. If you maintain mods, the autocomplete hook and prompt caching in $.model.complete are the new API surface to look at.