Environment variables, all 1,448.

OTEL_METRICS_EXPORTER ◇ Set OTEL_METRICS_EXPORTER=prometheus and child sessions report through the runner, giving one scrape target per machine.v2.1.138 or earlier → v2.1.308 ◆in the build OTEL_RESOURCE_ATTRIBUTES ◇ The gateway config accepts telemetry.resource_attributes, labels passed to spawned sessions when telemetry.forward_to is setv2.1.160 → v2.1.308 ◆in the build OTEL_SERVICE_NAME ◇ You can now set OTEL_SERVICE_NAME, OTEL_TRACES_EXPORT_INTERVAL and CLAUDE_CODE_FORK_SUBAGENT in settings.json.never mined ·never mined OTEL_SPAN_ATTRIBUTE_VALUE_LENGTH_LIMIT ◇ Maximum length of span attribute values in the bundled OpenTelemetry trace SDK; unset means no limit. Claude Code also uses the smallest of this, OTEL_ATTRIBUTE_VALUE_LENGTH_LIMIT, OTEL_LOGRECORD_ATTRIBUTE_VALUE_LENGTH_LIMIT and CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH (default 61440) as the character cap on the content it records in telemetry, such as prompts, system prompts, model output and tool inputs and results.v2.1.213 → v2.1.308 ◆in the build OTEL_TRACES_EXPORT_INTERVAL ◇ You can now set OTEL_SERVICE_NAME, OTEL_TRACES_EXPORT_INTERVAL and CLAUDE_CODE_FORK_SUBAGENT in settings.json.v2.1.138 or earlier → v2.1.308 ◆in the build OTEL_TRACES_EXPORTER ◇ Telemetry export can now send metrics, logs, and traces to different destinations instead of forcing all three to OTLPv2.1.138 or earlier → v2.1.308 ◆in the build P4PORT ◇ Any non-empty value adds perforce to the version control systems Claude Code reports in the vcs field of its telemetry environment metadata, alongside the ones it detects from marker files such as .p4config in the working directory.v2.1.138 or earlier → v2.1.308 ◆in the build PATH ◇ Inline admin policy scripts run with a stripped environment and a PowerShell resolved from fixed paths only.v2.1.138 or earlier → v2.1.308 ◆in the build PATHEXT ◇ About 220 environment variables now parse and validate in one place, including bounds checks.v2.1.138 or earlier → v2.1.308 ◆in the build PIP_CERT ◇ SSL certificate error message now lists the specific env var to check per toolv2.1.251 → v2.1.308 ◆in the build PIP_CONFIG_FILE ◇ No description from anybody.v2.1.242 → v2.1.263 ○gone PKG_CONFIG_PATH ◇ No description from anybody.v2.1.138 or earlier → v2.1.252 ○gone PLAYWRIGHT_BROWSERS_PATH ◇ Artifact preview now stays off when no Chrome, Chromium or Edge is found, unless CLAUDE_CODE_ARTIFACT_PREVIEW=true is setv2.1.160 → v2.1.308 ◆in the build PREFIX ◇ On Termux (when TERMUX_VERSION is also set), Claude Code puts its daemon directory (cc-daemon-<uid>) under $PREFIX/tmp instead of /tmp, along with its fallback SOCKS socket path (cc-socks-<uid>/<pid>.sock, used when the usual path under XDG_RUNTIME_DIR would be longer than 103 bytes). Separately, a non-empty value is treated like NPM_CONFIG_PREFIX: <PREFIX>/etc/npmrc joins a list of credential file paths, and a relative value makes that code throw an npm prefix variable holds a relative path.v2.1.138 or earlier → v2.1.308 ◆in the build PROGRAMDATA ◇ More environment variables, including Windows-specific ones, now pass through to Claude Codev2.1.282 → v2.1.308 ◆in the build PROJECT_DOMAIN ◇ Any non-empty value makes Claude Code report its deployment environment as glitch in telemetry environment metadata, unless an earlier-checked platform (such as Codespaces, Gitpod or Replit) already matched.v2.1.138 or earlier → v2.1.308 ◆in the build PWD ◇ PWD and OLDPWD now always count as risky variables when the Bash check decides whether rm or cd can take the quick pathv2.1.138 or earlier → v2.1.308 ◆in the build PYTHONSTARTUP ◇ No description from anybody.v2.1.242 → v2.1.263 ○gone RAILWAY_ENVIRONMENT_NAME ◇ A non-empty value, or a non-empty RAILWAY_SERVICE_NAME, makes Claude Code report its deployment environment as railway in telemetry environment metadata, unless an earlier-checked platform (such as Codespaces, Replit, Glitch or Vercel) already matched.v2.1.138 or earlier → v2.1.308 ◆in the build RAILWAY_SERVICE_NAME ◇ A non-empty value, or a non-empty RAILWAY_ENVIRONMENT_NAME, makes Claude Code report its deployment environment as railway in telemetry environment metadata, unless an earlier-checked platform (such as Codespaces, Replit, Glitch or Vercel) already matched.v2.1.138 or earlier → v2.1.308 ◆in the build RENDER ◇ When set to 1, true, yes or on, Claude Code reports its deployment environment as render in telemetry environment metadata, unless an earlier-checked platform (such as Codespaces, Vercel or Railway) already matched.v2.1.138 or earlier → v2.1.308 ◆in the build REPL_ID ◇ If REPL_ID or REPL_SLUG is non-empty, Claude Code detects its deployment environment as replit (checked after Codespaces, Gitpod, Coder, DevPod, Daytona, Google Cloud Workstations and AWS Cloud9). The result goes into the deploymentEnvironment field of the envContext metadata built for Claude Code's events; a separate environment-facts list also maps it to replit.v2.1.138 or earlier → v2.1.308 ◆in the build REPL_SLUG ◇ If REPL_SLUG or REPL_ID is non-empty, Claude Code detects its deployment environment as replit (checked after Codespaces, Gitpod, Coder, DevPod, Daytona, Google Cloud Workstations and AWS Cloud9). The result goes into the deploymentEnvironment field of the envContext metadata built for Claude Code's events; a separate environment-facts list also maps it to replit.v2.1.138 or earlier → v2.1.308 ◆in the build REQUESTS_CA_BUNDLE ◇ SSL certificate error message now lists the specific env var to check per toolv2.1.251 → v2.1.308 ◆in the build RUNNER_ENVIRONMENT ◇ CLAUDE_CODE_SUBAGENT_MODEL_FORCE joins a list of passed-on env vars, and RUNNER_ENVIRONMENT is added next to GITHUB_ACTIONSv2.1.138 or earlier → v2.1.308 ◆in the build RUNNER_OS ◇ When GITHUB_ACTIONS is truthy, the value of RUNNER_OS is recorded as githubActionsRunnerOs in the envContext metadata built for Claude Code's events. It is also read into the runnerOs field of a separate environment-facts object.v2.1.138 or earlier → v2.1.308 ◆in the build RUSTUP_HOME ◇ Protects rustup's directory from plugin writes: when a plugin's function hooks call $.fs.writeFile, a target at or under the path in RUSTUP_HOME is refused as a startup file or folder that rustup reads on its next start. An unset or empty value protects nothing.v2.1.242 → v2.1.263 ○gone SAFEUSER ◇ Read by the built-in /commit-push-pr command: the value, with every character outside a-zA-Z0-9._- removed, is listed as SAFEUSER in the prompt's context section next to whoami (from USER). The prompt tells the model to use it as the branch-name prefix when it creates a branch, falling back to whoami if it is empty.v2.1.138 or earlier → v2.1.308 ◆in the build SCREENSHOT_DIR ◇ No description from anybody.v2.1.145 → v2.1.245 ○gone SDK_NATIVE_BIN ◇ File name of the native binary that the bundled Agent SDK looks for inside the platform packages (@anthropic-ai/claude-agent-sdk-<platform>-<arch>, with -musl or -android variants on Linux), with .exe appended on Windows. The first candidate path that exists is used. Default claude when unset.v2.1.138 or earlier → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_AGENT_PROXY_URL ◇ The self-hosted runner reads its proxy from --agent-proxy-url, limits checkout hook time and tightens how nested-repository diffs are shownv2.1.307 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_BASE_DIR ◇ The self-hosted runner fails fast if its base directory is unwritable or the mount hangs.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_BG_RESULT_GRACE_MS ◇ Self-hosted runners hold shutdown until a finished background task is read; the grace window is configurable.never mined ·never mined SELF_HOSTED_RUNNER_CHECKOUT_HOOK_TIMEOUT_MS ◇ No description from anybody.v2.1.307 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_CLIENT_LABEL ◇ Name a self-hosted runner with --client-label or an env var instead of using the hostname.v2.1.248 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_CONFIGURE_GIT ◇ A truthy value (1, true, yes, on) turns on --configure-git for claude self-hosted-runner: at startup the runner writes global git config setting user.name to Claude, user.email to [email protected], SSH signing of commits and tags through a code-sign shim under <base-dir>/.runner/, and push.negotiate. A failure there is fatal; the setting also makes commits made by lifecycle hooks signed per session.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_CONFINE_REPO_SETTINGS ◇ The runner can warn about or block repo-committed settings that grant access outside the workspace.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DEBUG_DIR ◇ Directory for debug artifacts of the claude self-hosted-runner orchestrator: after each spawn-runner hook run it writes <jti>.jwt, <jti>.json (the decoded work-order claims) and <jti>.stderr (the hook's stderr or failure message) with mode 0600, and deletes matching files older than 5 minutes. The directory is created at startup; if that fails the orchestrator continues without debug artifacts.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DEBUG_TOKEN_DIR ◇ Directory where claude self-hosted-runner writes live credentials for debugging: runner_token.jwt at registration and on each refresh, and per session session_token_<id>.jwt and inference_token_<id>.txt. The directory is created with mode 0700 and the files with mode 0600; write failures are logged and ignored. --debug-token-dir sets the same option.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DEFER_SHUTDOWN_MAX_MS ◇ Self-hosted runners can keep serving attached sessions on the first shutdown signal instead of draining.v2.1.238 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DIFF_FINDS_NESTED_REPOSITORY ◇ The self-hosted runner reads its proxy from --agent-proxy-url, limits checkout hook time and tightens how nested-repository diffs are shownv2.1.307 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DRAIN_GRACE_MS ◇ CLAUDE_RUNNER_TRUST_CANONICAL_PREWARM now also needs a matching owner value, or the runner cleans up anywayv2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DRAIN_MARKER_FILE ◇ The self-hosted runner can now snapshot an operator's Claude config and seed it into every session, with a drain-marker mechanismv2.1.271 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_DRAIN_WAIT_MS ◇ Self-hosted runner adds git URL-rewriting flags and an Anthropic-managed git auth option, plus a documented drain-wait timeoutv2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_ENVIRONMENT_SECRET ◇ Runner "pool" flags are renamed to "environment"; old names still work but warn.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_EXEC_PATH ◇ Executable that claude self-hosted-runner spawns for child sessions. When unset or empty, the runner uses a command lifecycle hook from its hooks directory if one exists, otherwise its own executable. --exec-path sets the same option.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_HEALTH_PORT ◇ Port of the local HTTP listener of claude self-hosted-runner (and its orchestrator subcommand) that serves GET /healthz and GET /metrics. Default 8080; 0 disables it, and any value that is not an integer from 0 to 65535 is a startup error. A listener that fails to bind is logged and the runner continues without it.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_HOOKS_DIR ◇ The runner now has Windows errors for hooks and orchestrator mode, but they never fire.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_HOST_CONFIG_DIR ◇ The self-hosted runner can now snapshot an operator's Claude config and seed it into every session, with a drain-marker mechanismv2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_HOST_CONFIG_SNAPSHOT ◇ New SELF_HOSTED_RUNNER_HOST_CONFIG_SNAPSHOT env var configures self-hosted runner config storagev2.1.271 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_IDLE_SHUTDOWN_MS ◇ Milliseconds claude self-hosted-runner may sit without ever being assigned work before it exits for autoscaler scale-down. Unset or 0 means it never exits for idleness. The --exit-if-unused-min <n> option sets it from minutes.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_LOCK_TO_ACCOUNT ◇ Self-hosted runner errors about lock_to_account_id now point to --lock-to-account and say the value must be an organization memberv2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_LOG_FILE ◇ Path of a file that claude self-hosted-runner also writes its log lines to, opened in append mode with mode 0600; console output is unchanged. If writing fails, the runner logs a warning and continues without the file. An empty value counts as unset.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_MAX_LIFETIME_MS ◇ Maximum age in milliseconds of each session process spawned by claude self-hosted-runner; an unset, non-numeric or non-positive value disables it, and --kill-session-after-min sets it. At the deadline the runner normally releases the session (paused server-side, resumable) instead of killing it: one waiting on its user is released at once, and one with a turn in flight is released when that turn parks or finishes. A turn still running, or a release that is not accepted, is terminated after SELF_HOSTED_RUNNER_MAX_LIFETIME_GRACE_MS (default 15 minutes).v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_POOL_SECRET ◇ Runner "pool" flags are renamed to "environment"; old names still work but warn.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_POST_SESSION_HOOK_TIMEOUT_MS ◇ Timeout in milliseconds for the post-session lifecycle hook of claude self-hosted-runner, run at every session end. Default 60000; an unset, non-numeric or non-positive value uses the default. It is also added into the session kill grace and the runner's shutdown budget; --post-session-hook-timeout-sec sets it.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_PROXY_AUTHORIZATION_COMMAND ◇ The self-hosted runner can generate a fresh proxy auth token for every outbound connection.never mined ·never mined SELF_HOSTED_RUNNER_PROXY_AUTHORIZATION_FILE ◇ The self-hosted runner can generate a fresh proxy auth token for every outbound connection.never mined ·never mined SELF_HOSTED_RUNNER_PUSH_OUTCOME_ON_RELEASE ◇ Boolean (1, true, yes or on) equivalent of --push-outcome-on-release for claude self-hosted-runner. When on, a session that ends without completing (a shutdown drain, an idle release or a failure, but not a server-initiated deassign) has its tracked outcome branches pushed to the source remote, best-effort within a 30 second window, skipping branches with no new commits and repos checked out through the checkout lifecycle hook; a resumed session first fetches a previously pushed outcome branch and continues from it. The runner's shutdown budget grows by that push window, and startup warns when git is older than 2.29, which the resume fetch needs.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_REMOVE_SESSION_STATE ◇ Self-hosted runner can now delete a session's leftover files when the session endsv2.1.268 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_RETIRE_AT ◇ New flags retire a runner at a set time, release idle sessions, and control drain behaviour.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_SERVER_AUTO_MODE_LISTS ◇ A new SELF_HOSTED_RUNNER_SERVER_AUTO_MODE_LISTS setting lets self-hosted runner operators limit which server-supplied auto mode lists sessions followv2.1.295 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_SESSION_IDLE_MS ◇ Milliseconds a self-hosted runner session's user may be idle (or awaiting a user action) before the runner releases the session. --release-idle-session-min <minutes> sets it by writing minutes times 60000 into this variable. Unset, empty, non-numeric, 0 or negative disables idle release; values are capped at 2147483647.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_SESSION_STOP_GRACE_MS ◇ Milliseconds the self-hosted runner waits for the Claude process to exit cleanly after a session ends before force-killing it; it also feeds the shutdown budget the runner prints at startup. Default 5000; unset, empty, non-numeric, 0 or negative uses the default, and values are capped at 2147483647. --session-stop-grace-sec <seconds> sets it to the seconds times 1000.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_SIGKILL_TIMEOUT_MS ◇ Retired name for SELF_HOSTED_RUNNER_SESSION_STOP_GRACE_MS. If it is set to any value, including empty, the self-hosted runner logs a [runner:fatal] message saying it was renamed and exits with code 1 at startup.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_STARTUP_TIMEOUT_MS ◇ You can set a startup timeout for self-hosted runners, otherwise the built-in default applies.v2.1.224 → v2.1.308 ◆in the build SELF_HOSTED_RUNNER_TRUST_WORKSPACE ◇ Self-hosted runner switch, default on. While on, the runner seeds hasTrustDialogAccepted: true in the session's Claude config for each of the session's repo paths (and their resolved real paths) before starting the child. 0/false/no/off turns it off, unset or empty keeps the default, and any other value makes the runner throw an error; the --trust-workspace [bool] flag overrides it.v2.1.224 → v2.1.308 ◆in the build SESSION_INGRESS_URL ◇ A decoy tool tells the model browser tools already exist, only active in a specific remote session profile.v2.1.138 or earlier → v2.1.308 ◆in the build SESSIONNAME ◇ When SESSIONNAME is set and TERM is cygwin, Claude Code's terminal detection returns cygwin; that check comes after the Terminator, kitty, Alacritty, Tilix and Windows Terminal markers. Whether it is set is also recorded as hasSessionName in an environment report.v2.1.138 or earlier → v2.1.308 ◆in the build SHARP_FORCE_GLOBAL_LIBVIPS ◇ No description from anybody.v2.1.138 or earlier → v2.1.252 ○gone SHARP_IGNORE_GLOBAL_LIBVIPS ◇ No description from anybody.v2.1.138 or earlier → v2.1.252 ○gone SHELL ◇ NO_PROXY is now in the same list of environment variables as SHELL, TRACEPARENT and GIT_CONFIG_NOSYSTEMv2.1.138 or earlier → v2.1.308 ◆in the build SLASH_COMMAND_TOOL_CHAR_BUDGET ◇ Character budget for the skill listing Claude Code builds for the model, where each entry is a skill's name plus its description and whenToUse text (each capped at the skillListingMaxDescChars setting, default 1536). Any non-zero numeric value replaces the computed default, which is the context window size in tokens (200000 when none is passed) times a characters-per-token factor (default 4) times the skillListingBudgetFraction setting (default 0.01). When the listing is over budget, bundled skills keep their descriptions and the rest keep theirs in order of recent usage while they fit in what remains; the others, like skills set to name-only, are listed by name alone.v2.1.138 or earlier → v2.1.308 ◆in the build SPACE_CREATOR_USER_ID ◇ When set, Claude Code's hosting-environment detection returns huggingface-spaces; the check comes after the GCP, Azure and DigitalOcean markers and before GitHub Actions and GitLab CI. Whether it is set is also recorded under huggingface-spaces in an environment report.v2.1.138 or earlier → v2.1.308 ◆in the build SRT_DEBUG ◇ When set to any non-empty value, the sandbox runtime's debug logger prints its messages prefixed [SandboxDebug], warnings through console.warn and everything else through console.error. When unset, those messages are dropped.v2.1.138 or earlier → v2.1.308 ◆in the build SRT_WIN_PATH ◇ Path to the srt-win.exe helper that the sandbox runtime spawns. If set and the file exists it is used; otherwise Claude Code looks under the sandbox runtime's vendor/srt-win and vendor/srt-win-src build folders, and throws srt-win.exe not found when none exists.v2.1.169 → v2.1.207 ○gone SSH_AUTH_SOCK ◇ Whether the Chrome setup offer appears now depends on the app hosting Claude Code, not on SSH-related environment variablesv2.1.186 → v2.1.308 ◆in the build SSH_CLIENT ◇ More environment variables go through the central registry that handles allowlisting and scrubbing.v2.1.138 or earlier → v2.1.308 ◆in the build SSH_CONNECTION ◇ More environment variables go through the central registry that handles allowlisting and scrubbing.v2.1.138 or earlier → v2.1.308 ◆in the build SSH_TTY ◇ More environment variables go through the central registry that handles allowlisting and scrubbing.v2.1.138 or earlier → v2.1.308 ◆in the build SSL_CERT_DIR ◇ No description from anybody.v2.1.251 → v2.1.308 ◆in the build SSL_CERT_FILE ◇ Hosted cloud sessions can spawn a separate health-reporting helper, which never resolves on a normal install.v2.1.138 or earlier → v2.1.308 ◆in the build STY ◇ More environment variables go through the central registry that handles allowlisting and scrubbing.v2.1.138 or earlier → v2.1.308 ◆in the build SUDO_GID ◇ plugin eval now verifies a plugin's file tree is trustworthy before reading its manifestv2.1.235 → v2.1.308 ◆in the build SUDO_UID ◇ plugin eval now verifies a plugin's file tree is trustworthy before reading its manifestv2.1.235 → v2.1.308 ◆in the build SUDO_USER ◇ plugin eval now verifies a plugin's file tree is trustworthy before reading its manifestv2.1.235 → v2.1.308 ◆in the build SWE_BENCH_INSTANCE_ID ◇ Copied as sweBenchInstanceId into the metadata Claude Code attaches to its events, as an empty string when unset or empty. Also recorded as sweBenchInstanceId in an environment report.v2.1.138 or earlier → v2.1.308 ◆in the build SWE_BENCH_RUN_ID ◇ Copied as sweBenchRunId into the metadata Claude Code attaches to its events, as an empty string when unset or empty. Also recorded as sweBenchRunId in an environment report.v2.1.138 or earlier → v2.1.308 ◆in the build SWE_BENCH_TASK_ID ◇ Copied as sweBenchTaskId into the metadata Claude Code attaches to its events, as an empty string when unset or empty. Also recorded as sweBenchTaskId in an environment report.v2.1.138 or earlier → v2.1.308 ◆in the build SYSTEM_OIDCREQUESTURI ◇ No description from anybody.v2.1.138 or earlier → v2.1.259 ○gone SYSTEM_REMINDER_MEMORY_CONTEXT ◇ New SYSTEM_REMINDER_MEMORY_CONTEXT environment variable can force-enable background memory fetchesv2.1.267 → v2.1.308 ◆in the build SYSTEMROOT ◇ More Windows environment variables, such as SYSTEMROOT, COMSPEC and PSMODULEPATH, now get special handling when set in settings envv2.1.138 or earlier → v2.1.308 ◆in the build TASK_MAX_OUTPUT_LENGTH ◇ New settings let you control how much Bash and Task output Claude sees before it gets truncated to a file.v2.1.138 or earlier → v2.1.276 ○gone TEAM_MEMORY_SYNC_URL ◇ No description from anybody.v2.1.138 or earlier → v2.1.209 ○gone TEMP ◇ Working-tree upload now refuses linked worktrees and checkouts sitting in temporary scratch directories.v2.1.160 → v2.1.308 ◆in the build TERM ◇ Plugin eval helpers ignore machine-wide git settings so results cannot be skewed.v2.1.138 or earlier → v2.1.308 ◆in the build TERM_PROGRAM ◇ Environment reads go through one shared accessor; same variables, same behaviour.v2.1.138 or earlier → v2.1.308 ◆in the build TERM_PROGRAM_VERSION ◇ Read together with TERM_PROGRAM in several terminal capability checks: an iTerm.app major version of 3 or more gets truecolor output, tmux version 3.4 or later counts as supporting hyperlinks, and a ghostty version of at least 1.2.0 or an iTerm.app version of at least 3.6.6 turns on terminal progress reporting when the progressReporting setting is unset. It is also sent as term_program_version in the tengu_terminal_probe telemetry event.v2.1.138 or earlier → v2.1.308 ◆in the build TERMINAL ◇ On Linux, the command Claude Code prefers when it opens a new terminal window for a deep link: if set and it resolves to a command, that terminal is used. Otherwise it tries x-terminal-emulator, then a fixed list starting with ghostty, kitty and alacritty.v2.1.138 or earlier → v2.1.308 ◆in the build TERMINAL_EMULATOR ◇ The value JetBrains-JediTerm marks a JetBrains IDE terminal: terminal detection reports it as pycharm, hyperlinks are treated as supported, and the synchronized output check counts it as settled. Its raw value is also recorded in an environment detection payload as terminalEmulator.v2.1.138 or earlier → v2.1.308 ◆in the build