Jonathon's AI Wiki

What's New in Claude Code — Week 34 (August 17–21, 2026)

17 min read

Source: raw/anthropic-watch-claude-code-tag-v2-1-234.md (released 2026-08-17) and raw/anthropic-watch-claude-code-tag-v2-1-235.md (released 2026-08-18), both captured from the anthropics/claude-code releases feed on 2026-08-19.

v2.1.234 is one of the largest single releases the wiki has recorded — roughly sixty entries — and its centre of gravity is unattended operation: sessions that continue themselves when a usage limit resets, /goal that checks in on stalled background work instead of waiting forever, and permission dialogs that can be opened and changed mid-turn. v2.1.235 is a smaller consolidation release. v2.1.236–v2.1.239 (August 19–21) bring a built-in “Concise” output style, keybindingFlavor:"readline" customization, plugin headersHelper for dynamic marketplace auth, ANTHROPIC_DEFAULT_MODEL, /claude-api upgrade for Python migration, cost estimates with data-residency premium, and ListAgents now listing live teammates. Continues Week 33.

Key Takeaways

  • Claude Code now continues your session automatically when a claude.ai usage limit resets (v2.1.234). Off-switch lives in /config → “Continue automatically at usage limit.” This is the single most consequential entry of the week for anyone running long unattended work — the limit stops being a hard stop and becomes a pause. It also lands in the middle of an active community argument about limits (see Kahn v. Anthropic).
  • The claude-api skill’s context cost dropped from ~200k+ tokens to ~25k by loading its reference docs on demand (v2.1.234). An 8× reduction on a single built-in skill, achieved purely by deferring reference material — the largest concrete progressive-disclosure number the wiki has on file, and a direct measurement of the principle argued in New Rules of Context Engineering.
  • Claude is now instructed to use your account email only to identify you, and not to send it to unrelated services unless you ask (v2.1.234). A first-party data-handling constraint written into the model’s instructions rather than enforced by a tool boundary.
  • /goal stops hanging on dead or stalled turns (v2.1.234). It clears itself with a notice when a turn dies on an unrecoverable error (revoked auth, exhausted credit balance, context overflow) instead of staying armed; and when background tasks keep a goal waiting 30+ minutes, Claude checks in on them rather than waiting indefinitely. Opt out with CLAUDE_CODE_GOAL_CHECKIN_MINUTES=0.
  • Permission and configuration dialogs now open mid-turn. /permissions can be opened while Claude is working and rule changes apply to the rest of the current turn; /add-dir likewise; and /add-dir, /autocompact, /theme, /help, /config and /advisor all open mid-turn in the fullscreen TUI (v2.1.234). Previously a permission mistake meant interrupting the run.
  • Two permission-integrity fixes that were security-relevant, not cosmetic. Credential masking on relayed permission previews could previously hide commands, paths, or destinations from the approver — that is, the human was asked to approve something they could not fully see. Now oversized private-key blocks redact under full-strength redaction instead (v2.1.234). And Shift+Tab inside the permission prompt’s comment field approved the edit and granted session-wide edit permission rather than closing the field (v2.1.235).
  • The NT-namespace path hardening from W33 was extended to the remaining pre-approval file accesses (v2.1.234): remote file reads, session restore, CLAUDE.md includes, workflow scripts, and file uploads now all reject Windows \??\ paths. W33 closed this for UNC path validation; this closes the rest of the NTLM credential-leak vector.
  • Remote Control got materially more honest about state. Signing this computer into a different claude.ai account or organization now stops the running session within seconds with a reason, instead of surfacing a misleading HTTP 404 hours later. Sessions started from Desktop or VS Code keep phones and claude.ai/code updated on permission mode and model; effort picks made on a phone now apply to terminal- and Desktop-hosted sessions (v2.1.234).
  • SendMessage and ListAgents now say when your account’s session list was too long to check completely, instead of treating unseen sessions as absent (v2.1.234). A silent-wrong-answer failure mode turned into a visible one. SendMessage also refuses over-large messages up front rather than dropping them silently (v2.1.235).
  • The “Default teammate model” setting is gone from /config (v2.1.234). Agent-team teammates now use the leader’s model unless the spawn names one. If you relied on that setting to route teammates to a cheaper model, that routing must now be explicit at spawn time — see Agent Teams.
  • Todo/task list and GitLab work continued. A GitLab merge-request badge now appears in the footer and statusline (MR !N with draft/pending/green states) for repos with a GitLab remote and an authenticated glab CLI (v2.1.234), extending the GitLab push documented in W33.
  • Optional spellcheck in the prompt input (v2.1.235), underlining misspelled words as you type using an installed aspell, hunspell, or ispell.
  • Built-in “Concise” output style (v2.1.237): Claude leads with the result and skips preamble and narration — the zero-config way to reduce output verbosity without editing CLAUDE.md.
  • keybindingFlavor: "readline" (v2.1.238/v2.1.239): set in user or project settings to get Bash-style Ctrl+W, Alt+F, Alt+D, and punctuation-separated word keys in the prompt.
  • Plugin headersHelper (v2.1.238): marketplace definitions can now run a local command to mint fresh HTTP headers for catalog/archive fetches, enabling private authenticated marketplaces. Trust-dialog required.
  • ANTHROPIC_DEFAULT_MODEL (v2.1.236): new env var for setting the default startup model; /model picks still override and persist.
  • /claude-api upgrade (v2.1.239): migrates Python projects from anthropic 0.x to 1.x; the bundled skill’s Python reference updated for 1.x timeout API.
  • Cost estimates now include the 1.1× data-residency premium (v2.1.239). Enterprise teams using data-residency workspaces should re-check /cost and --max-budget-usd figures.
  • ListAgents lists live teammates and its own name (v2.1.239). Previously teammates appeared absent; now they show alongside subagents and other sessions. Sessions can now discover their own peer-addressable name.
  • Windows cross-session messaging (v2.1.239): SendMessage/ListAgents now available on Windows, matching macOS/Linux.
  • /goal check-ins now back off (v2.1.236/v2.1.239): 30 min → 1 h → 2 h, instead of repeating every 30 minutes. Goal restored on --resume.
  • notify_when_idle for SendMessage (v2.1.236): ask another session to notify you when it goes idle — opt-in, one-shot, no polling (macOS/Linux).
  • Custom output styles no longer drift mid-session (v2.1.238): a bug caused project/plugin output styles to revert to the default voice; now fixed.

v2.1.236–239: Notable additions (August 19–21)

Four more releases completed W34. The headline user-visible change is a first-party “Concise” output style; the headline infrastructure change is plugin headersHelper for dynamic auth; Windows gets cross-session messaging parity with macOS/Linux.

Built-in “Concise” output style (v2.1.237)

A new first-party output style is available under /config → Output style. “Concise” instructs Claude to lead with the result and skip preamble and narration, while doing the underlying work just as thoroughly. For users who want Claude Code’s output density reduced without authoring a custom output style or editing CLAUDE.md, this is the zero-config path. (v2.1.238 fixed a regression where custom, project, and plugin output styles were drifting back to the default voice mid-session — both fixes ship in the same week.)

keybindingFlavor: "readline" (v2.1.238)

A new keybindingFlavor setting (user or project scope) changes how Ctrl+W behaves in the prompt: "readline" makes it delete back to the previous whitespace, matching Bash muscle memory. The default "classic" is unchanged. v2.1.239 extended "readline" to cover the full Bash word-key set: Alt+F/Alt+D/Ctrl+→ for word navigation, and punctuation-separated word boundaries.

Plugin headersHelper for URL marketplace auth (v2.1.238)

Marketplace entries and url-type marketplace definitions can now specify a headersHelper — a local command that runs to mint HTTP headers (e.g. a short-lived auth token) for catalog and archive fetches. This enables private or authenticated plugin marketplaces without embedding credentials in config. headersHelper from a project .mcp.json or plugin now requires the folder’s trust dialog to have been accepted, and runs without inherited credential env vars for security isolation.

ANTHROPIC_DEFAULT_MODEL (v2.1.236)

A new ANTHROPIC_DEFAULT_MODEL environment variable sets the model new sessions start on. A /model pick still overrides it and persists across restarts. This is distinct from ANTHROPIC_MODEL, which does not persist through a /model override. Useful for CI environments or shared workstations where the preferred model should be set once.

notify_when_idle for cross-session SendMessage (v2.1.236)

SendMessage gained a notify_when_idle flag: ask another Claude Code session on this machine (macOS/Linux) to send you one notification when it next becomes idle. Opt-in, one-shot, no polling loop required. Useful for chained agent workflows where session B should start only after session A finishes a long task.

/claude-api upgrade (v2.1.239)

A new built-in skill command /claude-api upgrade migrates Python projects from anthropic 0.x to 1.x. The bundled skill’s Python reference was also updated: timeouts in the 1.x SDK use anthropic.Timeout, not httpx.Timeout. Part of the same v2.1.239 claude-api skill refresh that added Managed Agents Aug 19 features (web search/fetch domain settings, memory stores on self-hosted sandboxes) in v2.1.238.

Cost estimates include data-residency premium (v2.1.239)

/cost, the status line, and --max-budget-usd now include the 1.1× US-only-inference premium that applies to data-residency workspaces. Previously these estimates excluded the premium, so actual cost was higher than displayed. Enterprise teams using data-residency configurations should re-check budgets against the corrected estimates.

ListAgents lists live teammates + own name (v2.1.239)

Two ListAgents changes close gaps in agent-team visibility: it now lists your live teammates (previously only subagents and other sessions appeared, so a reachable teammate appeared absent), and it tells a session its own name (the one peers use to SendMessage to it). SendMessage to your own name now returns a clear message instead of “no agent named …”. ListAgents/SendMessage also no longer expose the idle pre-warmed worker the agent view pre-warms for the next background session — it now appears only once a task claims it.

Windows: cross-session messaging (v2.1.239)

SendMessage and ListAgents are now available on Windows, matching macOS and Linux. Claude Code sessions across Windows machines can now message each other and discover each other.

/goal backoff (v2.1.236 / v2.1.239)

The 30-minute /goal check-in introduced in v2.1.234 now backs off: after the first check-in at 30 minutes, subsequent ones are spaced at 1 hour, then every 2 hours. v2.1.239 also restores an active goal when resuming a session from the claude --resume picker.

Unattended operation is the theme

Three v2.1.234 entries point the same direction, and they are more interesting together than apart.

The usage limit stops being terminal. “Claude Code now continues your session automatically when a claude.ai usage limit reset” converts the most common hard stop in long-running work into a wait. Combined with /goal’s new 30-minute check-in on stalled background tasks and its self-clearing on unrecoverable errors, the failure modes that used to end an overnight run quietly now either recover or announce themselves.

Mid-turn control closes the other half of it. Being able to open /permissions and have rule changes apply to the rest of the current turn means a run that hits an unexpected permission wall no longer has to be killed and restarted. Same for /add-dir when the agent needs a directory nobody scoped in advance.

The honest caveat: auto-continue at limit reset is on by default, and it changes the cost profile of walking away from a session. If you deliberately used the usage limit as a budget stop, turn it off in /config.

The progressive-disclosure measurement

Reduced the context cost of loading the built-in claude-api skill from ~200k+ tokens to ~25k by loading reference docs on demand.

This deserves separating out because the wiki has argued the principle repeatedly without a first-party number attached to a specific artifact.

  • The reduction is ~8×, on a skill that ships with Claude Code.
  • Nothing was deleted — the reference docs still exist and still load, just on demand.
  • ~200k tokens is a substantial fraction of a context window spent before any work begins. The skill was, in effect, a worked example of the anti-pattern its own guidance warns about.

This is the same move the vault itself made when the schema was split into operations/ files, and the same one Agent Skills describes as the model for skill authorship. It also bears on a community argument that progressive disclosure still spends attention on the catalog — see Open Questions.

Selected fixes worth knowing

  • Auto mode in very long sessions was repeatedly re-checking and denying sandboxed commands’ network access after the conversation had been compacted (v2.1.234). If auto mode seemed to get more restrictive the longer a session ran, this was why.
  • Session-scoped permission answers, including denies, were dropped when answering background subagent tool permission prompts (v2.1.234). A deny you gave could silently fail to persist.
  • MCP diagnostics were printing resolved secrets (v2.1.234). Scope-conflict warnings now show the configured ${VAR} form; connection-failure details show only the server origin.
  • strictKnownMarketplaces allowlists accepted SCP-style git marketplace sources whose host differs from the one git would actually connect to (v2.1.234) — an allowlist that did not constrain what it appeared to constrain.
  • Whole-prompt-cache invalidation when a language server disconnected or reconnected mid-session (v2.1.235). A silent, expensive cache miss.
  • The Agent tool advertised a general-purpose default in sessions where that agent is unavailable (v2.1.235); an omitted subagent_type there now returns a clear error listing the available agents.
  • Repository detection mis-read the host of git remotes with unusual userinfo, producing links and repo-specific behavior for the wrong host (v2.1.234).

Remote Control: start sessions from your phone (W34, v2.1.234–239)

The official W34 digest page prominently features this as one of three headline items — the wiki’s changelog-based coverage captured some Remote Control changes (honest state reporting, effort picks from phone applying to terminal sessions) but missed this specific feature.

Any machine running claude remote-control now shows up as a device card at the top of the Code tab in the Claude app. The flow: start Remote Control on your machine, open the Code tab on your phone, tap the device card, pick a directory, and the session starts on your machine from your phone. This is distinct from the existing Remote Control feature that connected an already-running session — it enables starting a session remotely.

Remote Control is also out of research preview as of this release. It had been in research preview since launch. (Corrected 2026-09-29: this line said “since W16”, but the changelog already lists “Multiple fixes for Remote Control” in v2.1.101 on April 10, and the digest does not say which release in v2.1.234–239 shipped the device cards. See Claude Code Remote Control.)

# Start on the machine you want to reach
claude remote-control
# Then open Code tab on your phone → device card appears

v2.1.240–241: the quiet end of the week (August 22–23)

Both releases carry the same one-line note in the GitHub release feed — “Bug fixes and reliability improvements” — with no itemisation. Recorded so the version sequence in this wiki stays complete and so a later changelog fetch has something to attach detail to.

VersionDateRelease note
v2.1.2402026-08-22Bug fixes and reliability improvements
v2.1.2412026-08-23Bug fixes and reliability improvements

Week 35 opened with v2.1.243 (Aug 24, a changelog/feed chore) and v2.1.245 (Aug 25), the latter fixing a startup crash on Linux distributions shipping glibc 2.44 — named examples: Arch Linux, CachyOS, and Fedora Rawhide. v2.1.242 and v2.1.244 did not appear in the release feed. The rest of Week 35 (v2.1.246–v2.1.251, Aug 25–28) is covered in What’s New — Week 35; v2.1.243 and v2.1.245 are also noted in the CLI reference.

The Concise style, per Anthropic (2026-08-25 addition)

Reporting the built-in Concise output style from v2.1.237, The Neuron Daily states that Boris Cherny called it a quick band-aid while Anthropic works on a longer-term fix for recent output-quality complaints.

That reframes the entry above. A verbosity setting reads as a preference knob; a stopgap reads as an acknowledgment that the default register is a live problem with a real fix pending. ^[Second-hand — relayed by a newsletter, not quoted from a first-party post. The attribution is unconfirmed.] The field-report side of the same complaint is recorded in Claude Opus 5.

Related and shipped the same week: Anthropic staff are using an /eli5 skill — a topic explained as if you know nothing, rendered as an HTML artifact with big pictures and few words — now installable from the community marketplace:

claude plugin marketplace add anthropics/claude-plugins-community
claude plugin install eli5@claude-community

Per Thariq (Anthropic), the value is not brevity but comprehension before you attack a problem: /eli5 how does this module work, /eli5 why did we make this tradeoff, /eli5 what caused this incident.

Try It

  • Decide whether you want auto-continue at usage limit. /config → “Continue automatically at usage limit.” Leave it on for long unattended runs; turn it off if you were using the limit as a spend brake.
  • If you run agent teams, check your model routing. The “Default teammate model” setting is gone — name the model at spawn time or teammates inherit the leader’s.
  • Re-test any workflow that relied on a session-scoped deny inside a background subagent, since those denies were previously being dropped.
  • If you author skills, read the claude-api reduction as a template: move reference material behind on-demand loading rather than trimming content.
  • On Windows, update. The \??\ path hardening now covers remote file reads, session restore, CLAUDE.md includes, workflow scripts, and file uploads.
  • Try the “Concise” output style. /config → Output style → Concise. No CLAUDE.md edits required.
  • Set keybindingFlavor: "readline" if you use Bash and find Ctrl+W in Claude Code cuts too much text. No longer needed from v2.1.261 (2026-09-04): the word keys match Bash for everyone and keybindingFlavor has no effect.
  • If you run a private plugin marketplace, evaluate headersHelper to replace embedded credentials with on-demand token minting.
  • Enterprise data-residency teams: re-check your cost estimates. /cost and --max-budget-usd now include the 1.1× premium.
  • For chained agent workflows on macOS/Linux, use notify_when_idle in SendMessage instead of polling ListAgents in a loop.
  • Windows users: cross-session messaging is now available — no workaround needed.

Open Questions

  • What is the auto-continue behavior’s interaction with billing? The release note says the session continues when the limit resets; it does not say whether anything is queued, retried, or charged differently in the interim.
  • Does the claude-api skill’s ~8× reduction generalize? ~200k → ~25k is reported for one built-in skill with unusually large reference docs. No methodology or per-skill breakdown is published, so this is not a rate to expect elsewhere.
  • Is the “use your account email only to identify you” instruction enforced anywhere below the model? It is written as an instruction to Claude, which is a weaker guarantee than a tool-level or network-level block. The release note does not say.
  • v2.1.230 remains skipped, as noted in W33; v2.1.234 and v2.1.235 are consecutive.
  • Does the progressive-disclosure win answer the “context tax” objection? A community argument (raw/reddit-1vry6gk.md) holds that model attention should not serve as the retrieval index at all at 500+ skills, since the catalog itself grows. Deferring a skill’s reference docs shrinks the body, not the catalog entry — so the objection is narrowed by this change, not resolved by it.
  • Does keybindingFlavor: "readline" support a fully custom key map, or only the preset "readline" and "classic" flavors? The changelog does not name additional values.
  • What is headersHelper’s credential isolation model for user-scope marketplace entries? The changelog notes it “runs without inherited credential env vars” for project/plugin scope and from the Claude config dir for user/managed scope — but does not name which env vars are stripped.
  • Does the “Concise” style apply to all models or only some? No model restriction is named in the release note.
  • What was in v2.1.240 and v2.1.241? Both are “bug fixes and reliability improvements” with no itemisation in the release feed. The official changelog on code.claude.com may carry detail the GitHub release notes omit; not fetched this pass.
  • v2.1.242 and v2.1.244 are absent from the release feed, matching the earlier v2.1.213 and v2.1.230 pattern of skipped versions. No explanation is published.

Graph View

Backlinks