Claude Corner
Claude Corner
Short, practical Claude tips for developers. Each one covers a single task. A tip carries an affiliate link only when it is about deploying or running something.
Claude Code v2.1.287 n: Prefix Searches Agent Sessions Instead of Starting Them2 Oct 2026
In Claude Code agent view, every prompt you type at the bottom starts a new session, so retyping a word to find an old session creates a duplicate instead. Claude Code v2.1.287 (1 Oct 2026) adds an n: prefix that turns the same box into a filter on session …
Claude Code Caps Reply Prose So Tables and Code Stay Full Width1 Oct 2026
Claude Code v2.1.282 added maxProseWidth, a settings.json key that wraps paragraphs, headings, lists, and blockquotes at a column count you pick while tables and code blocks keep the full terminal width. It takes a whole number of columns, minimum 40, and Claude Code ignores any other value. The docs …
Claude Code 2.1.285 Sets Plugin and Bundled MCP Options From the Shell30 Sept 2026
Claude Code v2.1.285 adds claude plugin configure, which shows a plugin’s options and which are unset, or saves values piped into --values-stdin. claude plugin install --config now also takes <server>.<key>=<value> for a bundled .mcpb server’s own settings, and --values-stdin does not accept those keys. The plugin docs pages still …
Claude Code 2.1.284 Adds an Undocumented Second Compact as a Safety Net29 Sept 2026
Claude Code v2.1.284 added a fix: when a compacted request is still too long, it compacts once more and keeps less of the recent conversation. Only the changelog describes this second pass, and no documented setting or environment variable controls it, so treat it as a safety net, not …
Claude Code Rewind Beats Corrections but Misses Bash and Subagent Edits28 Sept 2026
Press Esc twice on an empty prompt, or run /rewind, to back out of a bad Claude Code turn instead of correcting it. Choose Restore code, Restore conversation, or both, knowing that files changed through Bash and most subagent edits are not in the checkpoint. Run v2.1.260 or later …
Claude Code Attribution False Makes Older CLIs Ignore the Whole Settings File25 Sept 2026
Confirm every Claude Code CLI that reads the file is on v2.1.281 or later before you set `”attribution”: false`. On a mixed-version team, hide the stamps with empty `commit` and `pr` strings plus `”sessionUrl”: false`. Move off `includeCoAuthoredBy`, which has been deprecated since `attribution` replaced it in v2.0.62. Claude …
Set Claude Code Deny Ask and Allow Rules Before the First Edit24 Sept 2026
Turn the risky task into a short in-scope / out-of-scope checklist in Claude Desktop before any edit. In Claude Code, open `/permissions` and set explicit allow, ask, and deny rules while still in plan mode. Confirm deny wins, then leave plan mode only for the files and commands on …
Claude Opus 5.5 Defaults to Medium Effort So Pin It After Switching23 Sept 2026
Opus 5.5 defaults to medium effort (Opus 5 defaulted to high), so a silent upgrade can feel softer than expected. Adaptive thinking stays on at every effort level on Opus 5.5; effort is the lever for depth, latency, and usage burn. After you change the model in Claude Desktop …
Claude Code v2.1.277 Can Keep Shared AGENTS.md When CLAUDE.md Exists22 Sept 2026
Stop losing shared `AGENTS.md` playbooks when a repo also has `CLAUDE.md` or `CLAUDE.local.md`. Set Project instructions to `claude-md-and-agents-md` in `/config` (or the matching `pluginConfigs` setting) so both files load. Pin Claude Code to `v2.1.277` or newer before relying on native `AGENTS.md` support. You keep one `AGENTS.md` for Codex and …
Claude Code v2.1.271 Lets Focused Subagents Skip CLAUDE.md With omitClaudeMd21 Sept 2026
Keep focused subagents from reloading the full user, project, and local CLAUDE.md stack into every worker turn. Set `omitClaudeMd: true` on agent frontmatter or `--agents` JSON so those workers skip instruction files while managed policy still loads. Pin Claude Code to `v2.1.271` or newer (prefer your current fleet pin …
Claude Projects let one coordinator fan work into parallel cloud threads19 Sept 2026
Brief one coordinator with a single goal, then let it fan work into parallel Claude Code cloud threads. Put durable decisions in shared project memory so later threads inherit them instead of relearning them from you. If you do not have the redesigned Projects UI yet, wait or stay …
Claude Corner: Terminal Missing Skills → Sync From claude.ai18 Sept 2026
Keep the skills and plugins you enable on claude.ai available in Claude Code terminal sessions signed in with the same account. Opt out deliberately with syncClaudeAiSkills / syncClaudeAiPlugins when a laptop must stay isolated from account packs. Pin Claude Code to v2.1.276 first so skill sync and corporate gateways do not collide with the advisor_20260301 …
Claude Corner: Blind Agent Edits → Plan Mode Before Disk Changes17 Sept 2026
Start Claude Code in plan mode (--permission-mode plan or Shift+Tab) so it explores and proposes without editing until you approve. Pin plan as the project default with permissions.defaultMode, then approve into Manual, accept-edits, or auto when the plan is ready. Use Desktop's Code-tab mode selector the same way when you are not in the terminal, …
Claude Corner: Claude Code Quota Lockout → Cap Effort and Burn Fewer Tokens17 Sept 2026
Cap team burn with maxEffortLevel (Claude Code v2.1.267+) so routine sessions cannot climb into xhigh / max by accident. Steer per session with /effort, --effort, or CLAUDE_CODE_EFFORT_LEVEL, and save defaults with effortLevel / modelSettings. Watch the shared seat window with /usage, pick a cheaper model for routine work, plan first, and keep tool dumps short. …
Claude Corner: Thinking Overspend → Set Adaptive Effort16 Sept 2026
Drop legacy budget_tokens. On Claude Opus 5, adaptive thinking is the default and manual thinking.type: "enabled" returns 400. Steer depth with output_config.effort (low / medium / high / xhigh / max). Keep that field off the thinking object. Bound cost with a large enough max_tokens (hard cap on thinking plus text) and read usage.output_tokens_details.thinking_tokens. Your …
Claude Corner: Uncited Doc Answers → Enable Citations Blocks15 Sept 2026
Turn on citations.enabled on every document block so Claude returns structured locations instead of freeform "according to the docs" prose. Prefer plain text or PDF for sentence chunking, or custom content blocks when you need RAG-chunk granularity. Do not combine citations with output_config.format JSON schema. The API returns 400 because citation spans cannot satisfy a …
Claude Corner: Force JSON Schema on Tool-Free Replies9 Sept 2026
Tip-only: lock a tool-free Messages reply to a JSON Schema with output_config.format so parsers stop stripping markdown fences.
Claude Corner: Pin Tool Choice for Stable Agent Loops8 Sept 2026
Tip-only: pin Messages API tool_choice (auto, any, tool, none) and disable_parallel_tool_use so agent loops get deterministic tool turns.
Claude Corner: Set Prompt-Cache Breakpoints (512-Token Min)7 Sept 2026
Tip-only: place Anthropic prompt-cache breakpoints on the static tools and system prefix; clear the 512-token floor on Opus 5-class models.