DevAgentRadar → Claude Code

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

Context

Assistant: Claude Code Proposal: Upgrade Claude Code to v2.1.311 so 200K-token sessions compact instead of demanding 1M credits Summary: v2.1.311 makes standard-context sessions compact and continue past 200K tokens instead of demanding 1M credits, adds self-hosted-runner --hold-idle-file and --max-hold-min, and fixes CLAUDE.md freezes, MCP interrupt delays, SubagentStop double-fires, and stale MEMORY.md after compaction. Primary source: https://github.com/anthropics/claude-code/releases/tag/v2.1.311

Why it matters

This release is mostly reliability, but several fixes change how you should run Claude Code. Standard-context sessions no longer abort with "Usage credits required for 1M context" when a request exceeds 200K tokens; Claude Code now compacts and continues, so do not switch to 1M context just to survive long chats. Self-hosted runners can stay idle while leftover host processes are still reported, via --hold-idle-file and --max-hold-min. CLAUDE.md files with a very long list item no longer freeze startup or Read/Edit. Interrupts now take effect while an MCP resource read is unanswered or a skill is still downloading. SubagentStop hooks fire once when a subagent finishes, not about twice a minute while it runs—remove debounce workarounds. After automatic compaction, Claude no longer sees stale MEMORY.md/CLAUDE.md if you edited them earlier in the session. Background-service errors now tell you to run claude daemon stop --any, which ends background sessions. DISABLE_COMPACT is honored by agent teammates and rewind no longer offers summarize-from/up-to-here. Do not infer the omitted 38 changes.

Suggested actions

1. Upgrade Claude Code to v2.1.311, then stay on standard context: sessions that exceed 200K tokens now compact and continue instead of stopping with "Usage credits required for 1M context".
2. Start self-hosted runners with --hold-idle-file <file> and --max-hold-min <minutes> so an idle session stays up while the host still reports processes the runner left running.
3. In every SubagentStop hook, remove debounce or ignore-duplicate logic added because the hook fired ~twice a minute per running subagent; v2.1.311 runs it once when the subagent finishes.
4. When you see "background service did not respond", run: claude daemon stop --any (do not use the disabled claude daemon restart). Then resume with --resume, --continue, or --session-id.
5. If DISABLE_COMPACT is set, keep it: agent teammates no longer compact anyway, and the rewind menu no longer offers "Summarize from here" or "Summarize up to here".

Config surfaces this release may change

• context and compaction settings (high confidence) — check your repo before applying
• MCP servers (high confidence) — check your repo before applying
• skills (high confidence) — check your repo before applying
• agent context files — check your repo before applying
• hooks — check your repo before applying
• subagent definitions — check your repo before applying
• background and headless runs — check your repo before applying
• settings files — check your repo before applying
• permission rules — check your repo before applying
• sandbox settings — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/claude-code-v2-1-311-upgrade-claude-code-to-v2-1-311-so-200k-token-sessi and mark Applied, Skipped, or Failed. Proposal id: 59b66788-39a1-489b-87fd-32e8697ab6b3

Your job

1. Restate the change in one sentence.
2. Propose a minimal plan for my repo (or a throwaway pilot).
3. Implement only what I approve; prefer small diffs and tests.
4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.