Follow Discord
Sweep 22 Oct 2026 · 04:32Z Build v2.1.308 527 read Stable v2.1.308 Latest v2.1.308 Next v2.1.308 Feeds RSS JSON llms.txt llms-full.txt Unofficial
Reference
One family

Feature flags, all 2,081.

GrowthBook gates the client evaluates on startup. Most of what ships is behind one of these before it is announced.

Feature flags2,081seen across 149 builds
In v2.1.308889the newest inventory on disk
Gone314in an earlier build, not in the newest
Described0have a line in Anthropic's documentation
No docs line2,081no line in the documentation we hold

Every feature flag, and where it lived

938 of 2,081 feature flags One tick per build on this box, oldest on the left. A gap is a build the feature flag was not in.
State
This account
Anonymous
Compiled default

A is the one account this site reads, N the anonymous baseline, D what the release compiled in. 1,143 of 2,081 gates have been read at all: a run only records the gates its own release wrote about. Claude Code evaluates its gates on Anthropic's servers, so a reading of off can be a rollout this account sits outside of, and none of it says anything about your install. How to read these.

○ not in v2.1.308, ◇ we found no documentation describing it.

Feature flagWhat it doesSpan v2.1.138 → v2.1.308 Status Reading
tengu_use_file_checkpoints ◇ No description from anybody.never mined ·never mined never read tengu_vast_puddle ◇ No description from anybody.v2.1.295 → v2.1.308 ◆in the build never read tengu_vast_sunset ◇ No description from anybody.v2.1.299 → v2.1.308 ◆in the build never read tengu_velvet_anchor ◇ No description from anybody.never mined ·never mined never read tengu_verified_vs_assumed ◇ Default false. When on, the # Doing tasks section of the system prompt gains the line When reporting results, be accurate about what you verified vs. what you assumed., which tells the model to distinguish what it confirmed by running a command or reading a file from what it believes but did not check, and not to assert assumptions as facts.v2.1.138 or earlier → v2.1.308 ◆in the build never read tengu_vertex_fallback_check ◇ No description from anybody.never mined ·never mined never read tengu_vertex_upgrade_check ◇ An analytics event, not a feature flag. Claude Code logs it at the start of a check for newer default models on Vertex, when the provider is vertex, CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST is unset, and at least one pinned model tier is found to check. It carries stale_tiers, the number of such tiers; each tier whose default model has a Vertex id is then probed and reported separately as tengu_vertex_probe_result.never mined ·never mined never read tengu_vinteuil_phrase ◇ No description from anybody.never mined ·never mined never read tengu_violin_ ◇ Remote sessions started from the desktop app now need a second flag as well.never mined ·never mined never read tengu_violin_plan_classifier ◇ A new flag will control whether plan mode gets classifier-based permission checksv2.1.267 → v2.1.308 ◆in the build never read tengu_virtual_pine ◇ No description from anybody.v2.1.295 → v2.1.303 ○gone never read tengu_vivid_crayon ◇ No description from anybody.v2.1.302 → v2.1.308 ◆in the build never read tengu_voice_circuit_breaker_tripped ◇ When voice input pauses itself, it now tells you the specific reason.never mined ·never mined never read tengu_warm_sunrise ◇ A find command containing -rm no longer counts as read-only, so Claude Code asks before running itv2.1.289 → v2.1.308 ◆in the build never read tengu_web_fetch_provenance_prompt ◇ WebFetch's permission logic is now a reusable function that can re-check permissions after a redirectnever mined ·never mined never read tengu_web_fetch_transport ◇ A new telemetry event records the outcome and duration of each raw WebFetch network requestnever mined ·never mined never read tengu_web_setup_push_offer ◇ A new dialog can offer to connect GitHub to claude.ai after a git push, not yet confirmed livev2.1.267 → v2.1.308 ◆in the build never read tengu_webfetch_deadline_ms ◇ WebFetch requests now enforce a timeout and fail with a clear deadline error if exceededv2.1.268 → v2.1.308 ◆in the build never read tengu_wif_user_oauth_refresh_token_cleared ◇ An analytics event, not a feature flag. Claude Code logs it, with no fields, after a workload identity federation token request fails with HTTP 400 or 401 and an invalid_grant body, and Claude Code removes the stored refresh_token from the cached credentials file because it is still the same token that was used.never mined ·never mined never read tengu_willow_prism ◇ Default false. Turns on the Agent tool's cost steer for pro subscribers when CLAUDE_CODE_AGENT_COST_STEER is neither a true nor a false value: the Agent tool prompt gains the paragraph Do not spawn agents unless the user asks. and drops the bullet about launching several agents in one message, and the initial agent listing omits its concurrency note.v2.1.138 or earlier → v2.1.139 ○gone never read tengu_willow_tern_applied ◇ A writing-style prompt block can be forced on with an environment variable; otherwise it stays off.never mined ·never mined never read tengu_witty_wand ◇ Model effort level can now be set per-model via a gated remote-configured tablev2.1.283 → v2.1.308 ◆in the build never read tengu_wobbly_fern ◇ If a request is too big, you're told whether dropping attachments or compacting would help.v2.1.229 → v2.1.308 ◆in the build never read tengu_wondrous_haven ◇ No description from anybody.v2.1.295 → v2.1.308 ◆in the build never read tengu_woolly_owl ◇ A remote bridge session's /clear command can now automatically dismiss a local dialog blocking itv2.1.265 → v2.1.308 ◆in the build never read tengu_worker_permission_mode_restore ◇ Resuming a session that was in plan mode now puts you back in plan mode.never mined ·never mined never read tengu_workflow_authoring_skill_autoload ◇ A bundled workflow-authoring skill loads itself when you mention workflows, and multi-agent runs need your ask.never mined ·never mined never read tengu_workflow_launched ◇ Internal renaming of the local-workflow launch tool's code, no behavior changenever mined ·never mined never read tengu_workflow_saved ◇ Your saved workflows can be listed and stored through a storage backend rather than plain files.never mined ·never mined never read tengu_worktree_detection ◇ An extra pass-through step was added to the internal chain that reports git worktree detection timingnever mined ·never mined never read tengu_worktree_kept ◇ Leaving a worktree warns about possible unpushed commits instead of claiming zero.never mined ·never mined never read tengu_worktree_removed ◇ Worktree cleanup aborts rather than deleting a tree whose state it cannot safely determine.never mined ·never mined never read tengu_worktree_resume_root_rejected ◇ Resuming into a missing worktree now warns you, and headless runs exit with an error instead of continuing.never mined ·never mined never read tengu_worktree_stale_lock_released ◇ An analytics event, not a feature flag. Claude Code logs it after a sweep that releases stale liveness locks on its own git worktrees, when at least one release was attempted. It carries attempted, the number of locks it tried to release in that sweep.never mined ·never mined never read tengu_woven_parcel ◇ No description from anybody.v2.1.305 → v2.1.306 ○gone never read tengu_write_tool_not_read_hypothetical ◇ Remotely served edits are less likely to be blocked by the read-before-write rule.never mined ·never mined never read tengu_year_end_2025_campaign_promo ◇ No description from anybody.never mined ·never mined never read tengu_zippy_harbor ◇ Six features controlled from Anthropic's servers are now checked early at startup, and five are off unless switched on remotelyv2.1.300 → v2.1.308 ◆in the build never read
Feedback