Environment variables, all 1,448.

CLAUDE_CODE_RETIRE_UNANSWERED_PARKED_PERMISSION ◇ Set an env var so resumed print-mode sessions drop unanswered permission requests instead of cancelling and re-asking.v2.1.233 → v2.1.274 ○gone CLAUDE_CODE_RETRY_WATCHDOG ◇ Persistent retry mode for API requests. When true (1, true, yes or on), 429 and overloaded errors are always retried and do not stop at the retry limit, repeated 529 errors do not raise the overloaded error, the default retry count is 300 instead of 10, and CLAUDE_CODE_MAX_RETRIES is not clamped to 15. The self-hosted runner sets it to 1 for the sessions it starts.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS ◇ Set CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS to make a request fail instead of waiting past that time on a usage limitv2.1.301 → v2.1.308 ◆in the build CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS ◇ Setting CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS now stops Claude Code from skipping a check it otherwise skips because of the retry watchdogv2.1.295 → v2.1.308 ◆in the build CLAUDE_CODE_RETRY_WATCHDOG_USAGE_LIMIT_RECHECK_MS ◇ No description from anybody.v2.1.305 → v2.1.308 ◆in the build CLAUDE_CODE_RIPPLING_TULIP ◇ A new CLAUDE_CODE_RIPPLING_TULIP value sets the token count for the total-tokens reminder, which can be per subagentv2.1.290 → v2.1.308 ◆in the build CLAUDE_CODE_RUSTLING_PIXEL ◇ Overrides the tengu_rustling_pixel flag (default all) that picks when Claude Code keeps thinking from a different model across a model change: all turns on api_keep_thinking_on_model_change, upgrade turns on api_keep_thinking_on_upgrade, none turns on neither. Any other value logs a warning and uses upgrade. The mode is none regardless when an earlier gate or the first-party API base URL check fails; _CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL skips the base URL check.v2.1.280 → v2.1.308 ◆in the build CLAUDE_CODE_SABLE_THRUSH ◇ Claude Code now spots server-tagged narration thinking, but the display change stays off by default.v2.1.242 → v2.1.250 ○gone CLAUDE_CODE_SAFE_MODE ◇ Setting CLAUDE_CODE_RESTRICT_PERSONAL_CONFIG starts a session that restricts your own Claude Code files, alongside the other restricted modesv2.1.169 → v2.1.308 ◆in the build CLAUDE_CODE_SANDBOX_LEAVE_DENIES_TO_SRT_WIN ◇ On Windows, the sandbox now installs and renews its file deny rules itself; set CLAUDE_CODE_SANDBOX_LEAVE_DENIES_TO_SRT_WIN to turn this offv2.1.299 → v2.1.308 ◆in the build CLAUDE_CODE_SANDBOX_LEAVE_PARENT_FOLDERS_ALONE ◇ On Windows, the sandbox grants itself access to the project's parent folders and the claude program, and will not start in your home or config folderv2.1.299 → v2.1.308 ◆in the build CLAUDE_CODE_SANDBOXED ◇ Four more environment variables are now recognized and passed through by Claude Codev2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SCRIPT_CAPS ◇ JSON object mapping a string to a numeric cap on how many times it may appear across the shell commands a session runs. It is read once, and applied only when CLAUDE_CODE_SUBPROCESS_ENV_SCRUB is true: before each shell command Claude Code counts each key's occurrences in the command text and fails with Script call limit exceeded once the running total passes the cap. Invalid JSON, a non-object, or no entries with a finite number disables it.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SCROLL_SPEED ◇ Scroll speed value for mouse wheel scrolling in the terminal UI. A positive number is used, capped at 20; unset, empty, non-numeric or <= 0 uses the terminal-dependent default of 3 or 1. The scroll speed settings dialog sets or clears it in Claude Code's own environment.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SDK_HAS_HOST_AUTH_REFRESH ◇ Host auth token refresh now returns the whole response, but only registers when an environment variable is set.v2.1.145 → v2.1.308 ◆in the build CLAUDE_CODE_SDK_HAS_OAUTH_REFRESH ◇ A host program can take over refreshing your login token, off unless an env var is set.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SDK_READS_SESSION_STATE ◇ The Agent SDK now ends a run when Claude Code reports it is idle, which can mean waiting up to 10 minutes after the resultv2.1.284 → v2.1.308 ◆in the build CLAUDE_CODE_SDK_READS_STARTUP_MARKS ◇ Setting CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS keeps only remote MCP servers from plugins, and several other environment variables are newv2.1.304 → v2.1.308 ◆in the build CLAUDE_CODE_SEND_FEEDBACK ◇ The switches hiding /feedback and /bug now behave the same in menus and commands.v2.1.212 → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_ ◇ Two env vars let background dispatchers inherit your plan and rate tier without an API call.never mined ·never mined CLAUDE_CODE_SESSION_ACCESS_TOKEN ◇ Remote settings can now supply many more environment variables, including auth tokens and home directory paths.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_ATTENDED ◇ A new CLAUDE_CODE_SESSION_ATTENDED environment variable tells child processes whether the parent session is interactivev2.1.269 → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_ID ◇ Skills from your claude.ai account can load as a source, but only if you set the sync env var.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_KIND ◇ You get a one-time warning line if a query names a model Claude Code does not recognise.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_LOG ◇ Path recorded as logPath in the session's entry in Claude Code's concurrent sessions registry (the per-process pid file).v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_NAME ◇ A blank CLAUDE_CODE_SESSION_NAME now falls back to the directory-derived name.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_ORIGIN ◇ A new CLAUDE_CODE_SESSION_ORIGIN environment variable is read and passed through to spawned subsessions.v2.1.259 → v2.1.308 ◆in the build CLAUDE_CODE_SESSION_START_ANNOUNCEMENTS_BEFORE_PROMPT ◇ A new env var lets session-start announcements be placed before the closing line in headless runsv2.1.277 → v2.1.308 ◆in the build CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS ◇ Nonsense values in retry, MCP output and hook timeout env vars now fall back to defaults.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SETTLE_TOUCHED_FILES ◇ Setting CLAUDE_CODE_SETTLE_TOUCHED_FILES can now turn on the check that skips unchanged files when Claude Code looks for editsv2.1.299 → v2.1.304 ○gone CLAUDE_CODE_SHELL ◇ Shell for Claude Code to run commands in, overriding shell detection. The path must contain bash or zsh and pass Claude Code's shell check; otherwise it logs is not a valid bash/zsh path, falling back to detection and detects a shell from SHELL and the usual install directories. Bash tool telemetry records whether it is set as executor_shell_overridden.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SHELL_PREFIX ◇ Hooks now run with the shell prefix and environment of whichever side made the call.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SHOJI_ENGINE ◇ No description from anybody.v2.1.193 → v2.1.202 ○gone CLAUDE_CODE_SILENT_TURN_REMINDER ◇ Claude can be nudged to check in when it has gone quiet for several turns, off by default.v2.1.236 → v2.1.308 ◆in the build CLAUDE_CODE_SILENT_TURN_REMINDER_SECONDS ◇ Set CLAUDE_CODE_SILENT_TURN_REMINDER_SECONDS to trigger the silent-turn reminder after a period of silence, not just turn countv2.1.290 → v2.1.308 ◆in the build CLAUDE_CODE_SILENT_TURN_REMINDER_TEXT ◇ Silent-turn reminder and artifact room environment variables are now recognised by name in the settings allowlist.v2.1.238 → v2.1.308 ◆in the build CLAUDE_CODE_SILENT_TURN_REMINDER_TURNS ◇ Claude can be nudged to check in when it has gone quiet for several turns, off by default.v2.1.236 → v2.1.308 ◆in the build CLAUDE_CODE_SIMPLE ◇ When a coordinator agent cannot use a tool, Claude is told to run it from a worker.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT ◇ Which models get the simple system prompt is now decided locally, not by remote config.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SIMULATE_PROXY_USAGE ◇ Claude Code can present thinking as connector text only, but nothing enables it here.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKILL_ATTRIBUTION ◇ A new environment variable lets you inject per-content-hash skill and plugin attribution into command and skill telemetry.v2.1.259 → v2.1.308 ◆in the build CLAUDE_CODE_SKILL_DESC_REFRAME ◇ Boolean. When true, the Skill tool's prompt uses the shorter Invoke a skill. text in place of the default Execute a skill within the main conversation text, and a debug line skill_desc_reframe_arm_active source=env is logged. When unset or false, the tengu_russet_linnet flag decides.v2.1.196 → v2.1.215 ○gone CLAUDE_CODE_SKILL_PROPOSALS ◇ An env var turns on the tool that proposes new skills; otherwise it stays off in this build.v2.1.233 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_ANTHROPIC_AWS_AUTH ◇ When true, the Claude Platform on AWS client (CLAUDE_CODE_USE_ANTHROPIC_AWS) does not look up AWS credentials: an Authorization header in the client's default headers (for example from ANTHROPIC_CUSTOM_HEADERS) is kept and also used as the API key with a leading Bearer stripped, and without such a header the client is created with skipAuth. At startup it also skips running the configured awsAuthRefresh and awsCredentialExport helpers, and the status panel shows Claude Platform on AWS auth skipped.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_ANTHROPIC_GOOGLE_CLOUD_AUTH ◇ More model providers are accepted, each with an environment variable to skip its auth step.v2.1.213 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_AWS_CRED_CACHE ◇ The gateway configuration accepts two new upstream providers, mantle for Bedrock Mantle and anthropicGoogleCloud for Google Cloudv2.1.207 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_BEDROCK_AUTH ◇ When true, Claude Code talks to Bedrock without looking up AWS credentials: its AWS SDK Bedrock clients use the smithy.api#noAuth auth scheme, and its Anthropic Bedrock client sends an Authorization header from the client's default headers (for example ANTHROPIC_CUSTOM_HEADERS) as its credential when one is present, or else is created with skipAuth; AWS_BEARER_TOKEN_BEDROCK still takes precedence. At startup it skips running the configured awsAuthRefresh and awsCredentialExport helpers, and the status panel shows AWS auth skipped.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS ◇ No description from anybody.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK ◇ Fast mode ('penguin mode') gains an admin setting to require per-session opt-in or disable it, plus an org-check bypass env varv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_FOUNDRY_AUTH ◇ For the Microsoft Foundry client, when neither ANTHROPIC_FOUNDRY_AUTH_TOKEN nor ANTHROPIC_FOUNDRY_API_KEY is set, a true value replaces the Azure DefaultAzureCredential token provider with the placeholder token skip-foundry-auth and keeps an Authorization header from the client's default headers if one is present. The status panel shows Microsoft Foundry auth skipped whenever it is true.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_HFI_VERSION_CHECK ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_MANTLE_AUTH ◇ More model providers are accepted, each with an environment variable to skip its auth step.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_MODEL_ACCESS_MEMORY ◇ Model access checks for AWS and Google Cloud accounts are now saved and reused at start-up, unless CLAUDE_CODE_SKIP_MODEL_ACCESS_MEMORY is setv2.1.285 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS ◇ Setting CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS keeps only remote MCP servers from plugins, and several other environment variables are newv2.1.304 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_PLUGIN_MCP_BUNDLES ◇ Setting CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS keeps only remote MCP servers from plugins, and several other environment variables are newv2.1.304 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_PLUGIN_MCP_SERVERS ◇ Plugin MCP servers that don't set alwaysLoad can now be deferred until needed in some cases, not only those with alwaysLoad set to falsev2.1.172 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_PLUGIN_MCP_SERVERS_EXCEPT ◇ Setting CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS keeps only remote MCP servers from plugins, and several other environment variables are newv2.1.199 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_PROJECT_BACKFILL ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_PROMPT_HISTORY ◇ Recalling an old prompt now restores its pasted attachments and remembers how you submitted it.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_REPO_UPLOAD ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_SKIP_VERTEX_AUTH ◇ A successful check that your cloud provider gives you access to a model is now saved on disk and reused for 15 minutesv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SLACK_TAG_TOKEN ◇ More Slack tokens and webhook URLs are detected and redacted, and one token is hidden from child processes.v2.1.234 → v2.1.252 ○gone CLAUDE_CODE_SLEEP_COMPACT ◇ Claude Code now reads CLAUDE_CODE_AUTOUPDATER_DISABLED_BY_HOST and CLAUDE_CODE_SLEEP_COMPACT at startupv2.1.295 → v2.1.308 ◆in the build CLAUDE_CODE_SLEEP_COMPACT_AFTER_MS ◇ Claude Code now reads three new environment variables: agent progress summaries, a sleep-compaction delay and a resume hook waitv2.1.296 → v2.1.308 ◆in the build CLAUDE_CODE_SLOW_OPERATION_THRESHOLD_MS ◇ No description from anybody.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SNAPPY_PELICAN ◇ Claude can now ask a subagent for lower or higher effort than its own, but the option is off unless switched onv2.1.296 → v2.1.308 ◆in the build CLAUDE_CODE_SPAWN_TIMESTAMP_MS ◇ Epoch time in milliseconds at which the process was spawned, parsed as a base 10 integer; an empty or non-numeric value is ignored, and CCR_SPAWN_TIMESTAMP_MS takes precedence when it parses. Claude Code uses it to report spawn_to_first_checkpoint_ms in the tengu_startup_perf event, as the start anchor (spawn_stamp) for time since session start when no session-switch time is known, and in remote sessions as the starting point of the auth cache warm-up clock.v2.1.152 → v2.1.308 ◆in the build CLAUDE_CODE_SPAWNED_BY_SERVER ◇ The Artifact tool's setting entry no longer names CLAUDE_CODE_DISABLE_ARTIFACT, though the variable name still appears elsewhere in Claude Codev2.1.307 → v2.1.308 ◆in the build CLAUDE_CODE_SQUISHY_NEWT ◇ Messages relayed from a project's coordinator session are handled by a switch that defaults to on, and the model is told when one is waitingv2.1.282 → v2.1.300 ○gone CLAUDE_CODE_SSE_PORT ◇ In some sessions Claude Code now ignores autoConnectIde and CLAUDE_CODE_AUTO_CONNECT_IDE=true when deciding whether to connect to your editorv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_STALL_TIMEOUT_MS_FOR_TESTING ◇ Milliseconds a streamed, checksummed file download (the one that checks a binary against a manifest size) may go without receiving data before it is aborted as stalled. Default 120000; an unset, 0 or non-numeric value uses the default. The error text still reads no data received for 120 seconds.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_STARTUP_FAILURE_RESULTS ◇ Version-mismatch startup failures now return structured JSON for SDK-driven sessionsv2.1.274 → v2.1.308 ◆in the build CLAUDE_CODE_STELLAR_DRIFT ◇ A new capability flag can be forced on with an env var, with no stated effect.v2.1.257 → v2.1.308 ◆in the build CLAUDE_CODE_STOP_HOOK_BLOCK_CAP ◇ How many consecutive times a hook may block a turn from ending before Claude Code overrides it: once the count exceeds the cap it logs tengu_stop_hook_block_count, shows the warning A hook blocked the turn from ending N consecutive times, and ends the turn. Default 8 when unset or non-numeric; 0 or a negative value disables the cap.v2.1.143 → v2.1.308 ◆in the build CLAUDE_CODE_STREAMED_BUMBLEBEE ◇ Setting CLAUDE_CODE_STREAMED_BUMBLEBEE_TEXT replaces the wording of the budget note given to subagents, with {N} standing for the budgetv2.1.290 → v2.1.308 ◆in the build CLAUDE_CODE_STREAMED_BUMBLEBEE_TEXT ◇ Setting CLAUDE_CODE_STREAMED_BUMBLEBEE_TEXT replaces the wording of the budget note given to subagents, with {N} standing for the budgetv2.1.290 → v2.1.308 ◆in the build CLAUDE_CODE_STRICT_INLINE_SHELL_RM_PROMPT ◇ Permission prompts for inline shell commands such as bash -c now depend on whether the command contains rm or rmdirv2.1.299 → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_CACHE_EVICT ◇ When true, turns on the prompt-caching-evict-2026-05-12 beta for model requests marked evictCacheOnComplete that carry sticky betas, without consulting the tengu_subagent_cache_evict flag (default false). Provider checks still apply: it has no effect unless the request goes to the first-party API at a first-party base URL (or Claude Platform on AWS with no ANTHROPIC_AWS_BASE_URL), and not when CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS is set or HIPAA mode is active.v2.1.198 → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_CONFIG_WARNING ◇ A startup notice can warn when your CLAUDE.md, skills or agents push Claude to delegate to subagents, which may cost extra tokensv2.1.295 → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_HISTORY ◇ Claude Code can list the subagents that ran earlier in a session, off unless CLAUDE_CODE_SUBAGENT_HISTORY is set or it is switched on remotelyv2.1.307 → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_MODEL ◇ Subagent model choice now records where the model came from, not just its name.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_MODEL_FORCE ◇ With the forced subagent model set, the workflow authoring reference drops its model-override guidance.v2.1.257 → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_PROMPT_CACHE_TTL ◇ You can pin the prompt cache to five minutes or an hour, separately for subagents.v2.1.242 → v2.1.308 ◆in the build CLAUDE_CODE_SUBAGENT_PROMPT_SNAPSHOT ◇ No description from anybody.v2.1.290 → v2.1.308 ◆in the build CLAUDE_CODE_SUBPROCESS_ENV_SCRUB ◇ A hidden --inherit-permission-mode flag lets a spawned session take its permission mode from whatever launched it.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SUBSCRIPTION_TYPE ◇ Subscription type attached to OAuth credentials supplied from outside: when the token comes from CLAUDE_CODE_OAUTH_TOKEN, CLAUDE_CODE_OAUTH_TOKEN_FILE_DESCRIPTOR or another externally supplied token, the credentials carry this value as subscriptionType, or null when it is unset or empty. Claude Code sets it itself when a background session consumes an auth snapshot and when it hands a token to a child process it launches, and removes it from the environment it passes to subprocesses.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SUPERVISED ◇ When true, an uncaught exception or unhandled promise rejection that Claude Code does not otherwise recover from or swallow is logged as Uncaught exception under CLAUDE_CODE_SUPERVISED (or Unhandled rejection under ...) and the process shuts down with exit code 70. After an unrecoverable interface error it also rethrows the error instead of exiting with code 1.v2.1.139 → v2.1.308 ◆in the build CLAUDE_CODE_SUPPRESS_SESSION_ATTRIBUTION ◇ When true (1, true, yes or on), Claude Code finds no session link to attribute: the remote or Remote Control session URL that would otherwise go into commit and PR attribution text as a Claude-Session: <url> line is left out. Same effect as setting attribution.sessionUrl to false.v2.1.161 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGIN_INSTALL ◇ Headless plugin installs refresh the cache in the background; set CLAUDE_CODE_SYNC_PLUGIN_INSTALL to skip it.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGIN_INSTALL_TIMEOUT_MS ◇ Plugin and MCP server syncing can now run alongside CLI startup under a time budgetv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGINS ◇ Org-synced plugin archives containing a .git entry are rejected and re-extracted safely.v2.1.157 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGINS_BUFFERED_DOWNLOAD ◇ When true (1, true, yes or on), plugin zips fetched by org plugin sync (/api/oauth/organizations/:orgUUID/plugins/<name>/download) are downloaded whole into memory, checked for a zip header and then written to disk. When unset they are streamed to disk under a stall watchdog (see CLAUDE_CODE_SYNC_PLUGINS_DOWNLOAD_STALL_MS).v2.1.179 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGINS_DOWNLOAD_STALL_MS ◇ Milliseconds a streamed org plugin sync download may go without receiving data before it is aborted with plugin download stream stalled and reported as a timeout. Default 60000; an unset, non-numeric or below-1 value uses the default. Has no effect when CLAUDE_CODE_SYNC_PLUGINS_BUFFERED_DOWNLOAD is on.v2.1.179 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGINS_INSTALL_TIMEOUT_MS ◇ Milliseconds Claude Code waits for org plugin sync installs, counted from when the sync starts during session setup, before carrying on without them (event tengu_plugins_sync_wait_timeout); late installs are picked up when they land. The same value caps the plugin reinstall on a reload_plugins control request and forwarded plugin installs. Default 30000; an unset, non-numeric or below-1 value uses the default.v2.1.157 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_PLUGINS_MCP_TIMEOUT_MS ◇ Milliseconds Claude Code waits for MCP servers from synced plugins to be reconciled after org plugin sync before moving on (event tengu_plugins_sync_mcp_timeout). Default 10000; 0 skips the wait (event tengu_plugins_sync_mcp_skipped) except while a relay hold is active, where a separate connect budget is used instead.v2.1.157 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_REUSE_WITHIN_MS ◇ Claude Code can skip a full skills sync on startup if one finished recentlyv2.1.268 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_SESSION_REFS ◇ Skills from your claude.ai account can load as a source, but only if you set the sync env var.v2.1.210 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_SKILLS ◇ Synced-skill state files are schema-checked, so one bad record no longer discards the lot.v2.1.149 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_SKILLS_INSTALL_TIMEOUT_MS ◇ Milliseconds a reload_skills control request waits for the skills sync reload to finish before it refreshes and returns the command list. Default 30000; an unset, non-numeric, 0 or negative value uses the default.v2.1.163 → v2.1.308 ◆in the build CLAUDE_CODE_SYNC_SKILLS_WAIT_TIMEOUT_MS ◇ Milliseconds, counted from when skills sync started, that a session waits for the first synced skills listing before refreshing its commands and continuing; on expiry it logs tengu_skills_sync_wait_timeout and goes on without waiting further. Default 5000; an unset, non-numeric, 0 or negative value uses the default.v2.1.149 → v2.1.308 ◆in the build