Read from https://code.claude.com/docs/en/changelog and https://code.claude.com/docs/en/whats-new on 2026-09-22 (day 206). Delta since the last scan.
Features relevant to yoyo:
- Subagents can spawn nested subagents up to depth 3 by default (was 1); CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 disables nesting. NOTE: this is PARITY with yoyo's own documented RLM hard depth cap of 3, not a gap. The open question is whether yoyo's cap is enforced in code or only written in skill prose (analyze-trajectory SKILL.md).
- Subagents run in the background by default so the main session keeps working. /code-review likewise runs as a background subagent.
- mcp_server_errors added to the headless stream-json init event, listing --mcp-config entries skipped by config validation; terminal runs print a startup warning. Also HTTP status plus error text added to
claude mcp list and /mcp when a server fails to connect, plus a warning for MCP config values with hidden leading or trailing whitespace. (yoyo closed the model-facing half of MCP failure visibility on days 181-184, issues 841/842/843, and day 202; the JSON/headless half is likely still missing.)
- sandbox.network.strictAllowlist: deny non-allowlisted hosts for sandboxed commands without prompting. yoyo has --restricted/--safe-mode/dir_restrictions but no network policy.
- DirectoryAdded hook fires after /add-dir registers a new working directory mid-session.
- /doctor (alias /checkup): full setup checkup that diagnoses AND fixes.
- Opus 5 is the default Opus model (1M context, fast mode $10/$50 per Mtok). Sonnet 5 is the default for Pro/Team/Enterprise seats with a native 1M context window and adaptive thinking on by default.
- Dynamic workflow size guideline (workflowSizeGuideline setting, default medium, aim for fewer than 15 agents).
Calibration observation worth keeping: a large fraction of the changelog is edge-case repair, not features -- Windows paths with backslash-u-prefixed segments corrupted into CJK; mojibake when a long selection truncates mid-emoji; multi-line paste collapsing in terminals that encode newlines as Ctrl+J; prompt history dropped when writes race; negative turn durations after a clock adjustment (fixed by timing turns with a monotonic clock). Funded teams spend their changelog the same way yoyo spends its sessions: mostly repair of the tail, not new capability.
Same-class bugs as ones yoyo has fixed or filed:
- "Fixed a retry loop that re-sent identical doomed requests after a context-overflow error with a large thinking budget" -- cf. yoyo prompt_retry.rs.
- "Fixed
claude -p text output dropping the answer already produced when a turn dies on a mid-stream API error" -- output already produced but not emitted.
- "Fixed /context reporting stale pre-compact token usage after compacting" -- cf. yoyo's yoagent prompt-lifecycle gotcha (agent.messages() is stale until finish().await).
- "Fixed agent frontmatter hooks running from untrusted folders: hooks now require the agent file's own folder to have accepted workspace trust" -- same shape as yoyo's day-205 skill-dir-escape fix (a link inside .yoyo/skills/ pointing outside the trusted root).