Blog

Amp build 0.0.1791676897 lets you run Amp on your own Claude Pro or Max subscription, which Sourcegraph lists as free and available to everyone. A second change gives Puck multiple separate conversations, each with its own Puck. Both landed in the week after the 0.0.1791060199 build.

OpenCode v1.19.0 adds opencode upgrade --major, an opt-in jump from the 1.x line to OpenCode 2, and stops the TUI from installing updates automatically (you run /update now). It also validates PTY connection origins and tickets, accepts OpenAI's ultrafast service tier, serves usage stats as JSON and Markdown, and fixes xAI tool-result images.

Codex CLI v0.162.1 is a patch on top of v0.162.0 with two reliability fixes. A multi-line asynchronous question no longer crashes the terminal UI, and line breaks and full hyperlink destinations are preserved. Startup no longer fails when a running background server's feature settings differ from the CLI's defaults, since the compatibility check now applies only to explicit command-line feature overrides.

Codex CLI v0.162.0 adds managed Git worktree tools behind the worktrees feature flag, a /copy command and Ctrl+Insert for copying transcript text, and clickable URLs across approval and MCP prompts. It adds opt-in ranked tool search in Code Mode, protects a new Guardian decisions key from environment forwarding, and introduces a codex-cloud-client crate for resuming and attaching to cloud threads.

OpenCode 2 v2.0.25 adds in-app previews for Word, Excel and PowerPoint files and stops asking for access to directories outside the project by default. The CLI now ships its browser UI as brotli-compressed raw bytes, pairing QR codes encode every address, and the server streams fs.read with HTTP Range support. Provider work adds between_tools thinking in Anthropic Messages plus missing OpenAI and Google service tiers.

OpenCode 2 v2.0.26 adds a Google Vertex auth path that reads your local gcloud Application Default Credentials instead of an API key (#53798), and moves remote access to a randomly generated, redacted subdomain so the tunnel address stays unguessable (#53962). Terminal links now open from any row of a wrapped link, untracked-file diffs are batched instead of one git process per file, and the publish pipeline starts writing reviewed release notes into CHANGELOG.md.

Claude Code v2.1.293 adds Claude Haiku 5.5 (`claude-haiku-5-5`), a 1M-context model priced at $0.10 input and $0.50 output per million tokens and set as the default Haiku on the Anthropic API. Subagent status-line scripts now receive `agentType` in their payload, and mods can set `isDeferred` on `$.tool.register` to list a tool's schema in the prompt instead of behind tool search. Fixes stop Claude retracting finished work after a compaction, close an HTTP MCP memory leak, and keep a queued message from being lost when the left arrow backgrounds a session.

Codex CLI v0.161.0 adds a /mcp login <name> command for signing into MCP servers from an active terminal session, with enterprise sign-in that cleans up account-scoped grants and fails closed on config refresh. It gates the Daybreak interface behind an opt-in cli_daybreak flag, adds a per-turn codex exec --cyber-access-program override, lets voice pick the microphone input channels, makes GPT-6.1 Sol the default catalog model, and ships sandbox, permission, and SQLite-recovery fixes.

Claude Code v2.1.290 stops WebFetch from silently dropping page text past 100,000 characters: it reports the unread amount and takes an offset to keep reading. Scheduled tasks set with /loop or reminders now survive resume, compaction, and a move to the background. New environment variables set the hourly web-search budget refill rate, the idle-compaction token threshold, and a proactivity off switch, and plugin tool.check hooks gain an organization approval ceiling.

Claude Code v2.1.291 is a patch over v2.1.290 whose release notes list two regression fixes: cloud sessions could drop your answers to permission prompts, and a session's last messages could be lost when quitting. Both are now resolved.

Claude Code v2.1.292 adds an effort parameter to the Agent tool so a sub-agent runs at the level you ask for, and a --marketplace flag that registers a marketplace and installs a plugin from it in one command. CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS lengthens backoff on overloaded (529) requests, and a permission-prompt bypass for file reads from network (UNC) paths is closed, alongside fixes for reads that followed a link swapped in mid-read.

Gemini CLI v0.63.0 is a reliability release with no new models. Plans now run to completion in non-interactive mode, an auth fix stops the infinite login loop from file contention, a headless keyring, and dropped supervisor state, and the CLI shows retry progress during connection recovery. A new GEMINI_CLI_AUTH_OVERRIDE variable carries the chosen auth type across the CLI's self-relaunch so the selection is not lost.

OpenCode 2 v2.0.24 adds Vercel AI Gateway as a native provider, with its language models in the catalog, and installs the AI SDK v6 provider packages by default. The catalog drops non-text models, routes completions-shaped Azure entries through chat, and resolves reasoning variants for GitLab Duo. OpenAI sign-in methods are relabeled Codex, and the terminal interface surfaces sessions awaiting input at the top of the open menu.

OpenCode 2 v2.0.23 adds a native Cohere chat provider and fixes system-part handling for Gemini, Mistral, and Cohere. A dense pass over the Agent Client Protocol improves running the agent inside editors like Zed, JetBrains, and Neovim. A new skill frontmatter field, disable-model-invocation, hides a skill from the model's auto-invocation list while keeping it loadable by ID. No new environment variables or experimental flags.

Amp build 0.0.1791060199 updates the models behind the Dial. Claude Opus 5.5 now powers the medium mode, which Amp says solves more tasks for less and is easier to steer. GPT-6 Astra inference runs six times faster. Both sit behind the Dial's tiers, not a model picker.

Claude Code v2.1.289 makes deny and ask rules hold where they used to slip: a nested part of a compound shell command, a Bash command behind an environment-variable prefix like TZ="$HOME" rm -rf build, and a file reached through a symlink. For plugin authors it adds an agent.spawn hook for teammates, idle and waiting states in $.agent.list(), and a ui.fault a failing Client raises so one mod's failure stays contained.

Claude Code v2.1.288 restores a prompt cleared with Ctrl+C when you press Up, prompts before a dangerous rm hidden in a bash -c script under bypass or shell-allow rules, and adds CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS for gateways that reject structured outputs. /code-review gains --max-findings, the agents view gets Ctrl+F and group jumps, and a cluster of fixes keeps long and resumed sessions alive.

OpenCode 2 v2.0.22 extends its Agent Client Protocol support so an editor like Zed gets additional workspace directories, model and agent changes that follow from other connected clients, and V2 forms answered by elicitation. Plugins gain session removal, compaction, and parent-session creation. Provider failures across DeepInfra, Novita, Together, xAI, and Z.ai now classify context overflow and auth errors instead of returning generic failures, and Azure deployment discovery plus prompt caching reach more providers.

Claude Code v2.1.287 adds Claude Mods, letting plugins register render sites, interfaces, and hooks that change deeper session behavior, plus an opt-in You-should-know side agent that flags things you or Claude might miss. MCP servers on the 2025-11-25 protocol can send URL sign-in prompts, with bareElicitationCapability as the switch for older server SDKs. A plugin directory, telemetry's prompt_text field, and a restored dangerous-rm safeguard round it out.

Codex CLI v0.160.0 lets sessions start outside a project directory with workspace defaults, gives Guardian reviews opt-in access to earlier user instructions and agent-handoff context, and paginates older tasks in the command center. Reliability fixes stop duplicate sends of queued messages after reconnect, harden the Windows sandbox, and keep the local SQLite session store from stalling.

OpenCode 2 v2.0.21 concentrates on provider routing and session UI. Cloudflare AI Gateway Claude and OpenAI calls route through native passthroughs, OpenRouter Anthropic and Qwen requests carry prompt cache breakpoints, and over-limit responses from Workers AI and Amazon Nova are classified as context overflow. The session view groups adjacent file reads into one row, keeps the open group header sticky, and opens session IDs referenced in markdown. No environment-variable or experimental-flag surface changed.

Claude Code v2.1.286 retries once on the previous model of the same tier when the API refuses your default model, so a refused model no longer fails every turn. Stacked permission requests show a "2 of 5" count, a shared lock stops several processes from each opening a login browser when GCP or AWS credentials expire, and a batch of redaction fixes keeps secrets out of error messages, logs, and the /feedback zip.

Codex CLI v0.159.3 rolls up two backports onto the 0.159 line since v0.159.1. v0.159.2 suppresses the console window that flashed on Windows when Codex launched background and sandboxed commands. v0.159.3 lets eligible local sessions signed in with ChatGPT show optional reminders to complete account security setup.

OpenCode v1.18.34 is a maintenance release over v1.18.33 with three Core bugfixes. macOS release binaries are now signed with a Developer ID, locally compiled macOS binaries are re-signed to run on macOS 27 and newer, and model requests carry namespaced session and parent-session identity headers. No environment variables or experimental config keys changed.

Claude Code v2.1.285 adds CLAUDE_CODE_DISABLE_WEB_FETCH to turn the WebFetch tool off, an allowedProviders managed setting that limits which API providers a machine may use, and claude plugin configure for reading and setting a plugin's options from the command line. CLAUDE_CODE_NONSTREAMING_TIMEOUT_RETRIES caps re-sends of a timed-out non-streaming request. It follows v2.1.284 with a large batch of request-path and Remote Control reliability fixes.

Codex CLI v0.159.0 adds an opt-in instant_interrupt that lets new input steer a turn mid-response, a compact welcome screen with in-turn tips, and a keep action in the warnings viewer. Windows launches stop flashing stray console windows for MCP servers and piped commands, transcript copies keep Markdown tables and whitespace, and network-enabled sandboxes get macOS TLS trust plus upstream proxying of permitted private IPs.

Codex CLI v0.159.1 is a patch release on the 0.159 line that promotes GPT-6.1 Sol to the default model in the bundled catalog. It also adds GPT-6.1 Sol to Amazon Bedrock's Mantle and Runtime catalogs, so teams running Codex through AWS get the new default on the same release.

Gemini CLI v0.62.0 follows v0.61.0 with two new models, Gemini 3.8 Flash and 3.5 Flash Lite (PR #29443), plus a round of reliability fixes. It retains the stored OAuth refresh token when a refresh response comes back partial, so long sessions stop hitting surprise re-authentication prompts. It also hardens Windows ConPTY and PTY lifecycle handling, guards the terminal-border renderer against negative and non-finite dimensions, and cleans up ACP and MCP tool-call formatting.

OpenCode 2 v2.0.19 fits each request's output limit to the context window and caps it at 256k tokens. A new OPENCODE_SESSION_ID environment variable is set in every command the shell tool runs. Media routes gain an ElevenLabs Scribe transcription path, and repository host references now reject relative path segments.

OpenCode 2 v2.0.20 adds `auth export` and `auth import` commands that move provider credentials as JSON between machines, and a way to disable the background service. Signing in through ChatGPT now registers a separate token-sharing OAuth method and uses those tokens for OpenAI models, with a usage-limit dialog and errors that explain token-sharing failures. Database files are restricted to their owner, and denied permissions no longer stop a non-interactive `run`.

Claude Code v2.1.284 adds Claude Sonnet 5.5 as the default Sonnet model on the Anthropic API (1M context, $2/$10 per Mtok). Interactive terminal and VS Code sessions now start in auto mode on every plan and provider, auto mode's outside-directory read prompt gains a 'Yes, but ask again next time' answer, and Ultracode becomes its own /effort toggle instead of forcing xhigh effort.

Codex CLI v0.158.0 adds configurable copy-on-select and Markdown-preserving copy in the fullscreen terminal, reaches MCP servers with pre-registered OAuth client secrets via codex mcp add --oauth-client-secret, and secures direct exec-server WebSocket connections with bearer tokens. Under the hood, WebSocket auth moved into a codex-websocket-auth component and command spawning now runs through a shared child-process launcher with a reaper. Sandbox fixes land on Windows, Linux, and macOS.

OpenCode v1.18.33 is a bugfix point release with no new environment variables. Debug configuration output now redacts credentials and sensitive headers, Cloudflare AI Gateway models honor provider response and stream timeouts, MCP browser launch failures get reported, and Gemini thinking defaults match each model generation's supported controls.

Eighteen point releases carried OpenCode 2 from v2.0.0 to v2.0.18 in a single week. The span builds out Code Mode, where the model calls tools by writing JavaScript in a sandbox, and adds image, video, speech, and transcription generation across providers. Two new env vars ship: OPENCODE_CLI_CONFIG_CONTENT for inline config and OPENCODE_GITLAB_AUTH_CLIENT_ID for GitLab OAuth. New experimental keys add a server fs.write endpoint, WebSocket plugin hooks, MCP lifecycle control, and session export and import.

Build 0.0.1790452833 adds runners that execute Amp threads on a machine you own, ending with --share runners a whole workspace can reach. Amp also went free when you bring your own compute and model keys, and Fable 5.1 now powers a new ultra mode on the Dial.

Claude Code v2.1.283 adds admin model governance: availableModelsMatch "exact" allows only the model versions you name so new releases stay blocked until listed, and deniedModels blocks specific models even when the allowlist permits them. Requests can carry an x-claude-code-prompt-id header (opt in with CLAUDE_CODE_GATEWAY_HINT_HEADERS=1) so an LLM gateway groups the calls behind one prompt, and /doctor prompt-audit flags CLAUDE.md files and prompts written for older models.

Coming from v0.153.4, v0.157.1 brings voice conversations, fullscreen transcripts by default, worktree-isolated sessions, and a background server with /daemon and --no-daemon controls. GPT-6 Astra, Sol, and Luna are in the model picker, with Sol and Luna adding Amazon Bedrock support in v0.157.0. OPENAI_FILE_BLOB_UPLOAD_TIMEOUT raised the file-upload default from 60 seconds to 5 minutes.

Between v0.53.1 and v0.61.0, Gemini CLI's eight stable releases went almost entirely into security. A new GEMINI_RESTRICTED_MODE fails workspace trust closed and filters out MCP servers, GOOGLE_CLOUD_WORKSTATIONS adds Cloud Workstations OAuth, and the MCP OAuth path gains SSRF blocking and RFC 9207 issuer checks. Sandbox and extension isolation tightened across the window; no new production model IDs shipped.

Every OpenCode 2 release analyzed: environment variables, config keys, and the features and fixes each 2.x build added, grounded in the tagged open-source releases and the OpenCode 2 docs.

Nineteen releases since v1.18.13 went into making OpenCode's multi-provider path reliable: automatic retries for more network and mid-stream errors, five-minute default timeouts for slow model startups, and resumable subagent failures. Azure now signs in through the Azure CLI with Microsoft Entra ID instead of an API key, and Cloudflare AI Gateway gets native OpenAI and Anthropic passthroughs. The work is in provider handling and the desktop client.

v2.1.269 adds `claude plugin eval` for scored, reproducible plugin test reports, folds a diff of the files a Bash command changed into the Bash tool result, and lets you raise the Workflow tool's concurrent-agent limit for large fan-outs. The 267-to-269 window also restores prompt-cache reuse that broke on resume after an interrupt and on the turn after an output-token cutoff.

Since v2.1.252, Claude Code makes Claude Fable 5.1 the default Fable model (1M context, $10/$50 per Mtok) and adds a fullscreen /diff panel that tracks uncommitted changes as Claude edits. Auto mode now asks before reading outside the working directory or an artifact you don't own, and v2.1.266 itself restores CLAUDE_CODE_USE_GATEWAY to being ignored when set alone.

Between v0.152.0 and v0.153.4, Codex CLI rolled GPT-6-Astra from an API-only option into the bundled default model over four point releases. v0.153.0 added Vim undo and redo, remote-marketplace install and removal in the plugin CLI, an optional automatic recap, and an experimental token-budget context mode for ChatGPT-plan sessions.

Across 0.151.0 and 0.152.0, Codex CLI added Vim / and ? search inside the composer, rate-limit banners that link straight to usage, credits, and plan actions, and per-tool MCP output_token_limit truncation that survives session resumes. The terminal and codex exec now show credential-refresh progress including Amazon Bedrock reauth, MCP server names accept package-style characters, and the planning tool is off by default (tools.update_plan.enabled = true). Beneath those, the range hardens the Guardian approval-review path and sandbox correctness across macOS, Windows, and remote executors.

v2.1.252 adds PreModelSwitch and PostModelSwitch hook events and closes several ways a permission check could be sidestepped: Read/Write/Edit following a symlink swapped after approval, Grep/Glob ignoring Read deny rules through symlinked paths, and plugin and Workflow paths reaching outside their scope. /cost gains a per-session prompt-cache line, /usage gains a spend-limit bar, and four new environment variables tune the served model catalog, remote-tools serving, MCP task backgrounding, and background-worker check-in cadence.

Between v0.146.0 and v0.150.1, Codex CLI added an interactive `codex agents` dashboard, `codex queue`, and `@`-task mentions for running and messaging many agent tasks at once. The range also brings scriptable and Interrupt hooks, opt-in MCP 2026-07-28 protocol support via CODEX_MCP_PROTOCOL_VERSION, a built-in Amazon Bedrock provider with GPT-5.6 routing, and session export/fork/archive.

v2.1.247 adds a /claude-api cost-optimize skill that profiles a project's Claude API spend and works the cost levers one measured change at a time, plus Admin API doc coverage. Organizations can now rotate spinner tips from a JSON file via spinnerTipsOverride.tipsFile, and Sonnet 5's 1M sessions auto-compact later (~967K). The changelog announces the SendFeedback drafting tool, but it already shipped in the v2.1.245 baseline.

v2.1.245 fixes a startup crash on Linux distributions shipping glibc 2.44 (Arch, CachyOS, Fedora Rawhide). The v2.1.243 window in this span adds per-session prompt-cache TTL control via promptCacheTtl/subagentPromptCacheTtl and the CLAUDE_CODE_PROMPT_CACHE_TTL override, a bounded first-byte timeout so a stalled API request fails fast instead of hanging, and a shared workload-identity-federation credential cache for CI.

The span from v2.1.238 adds the ability to delete an Artifact (owner-only, confirmation required, blocked in plan mode) and a lower-priority mode that keeps a Max session running after it hits the weekly limit. Both shipped in v2.1.239 without a changelog line, and a one-time nudge to set default effort to medium arrived in v2.1.240 under 'bug fixes.' v2.1.239 also fixes a Bedrock proxy bug that silently doubled billed API calls.

Spanning v2.1.235 through v2.1.237, this release adds a built-in Concise output style that drops narration while keeping error and safety output, the ANTHROPIC_DEFAULT_MODEL variable that pins the starting model without blocking a /model switch, and notify_when_idle on SendMessage. The build also carries an unannounced silent turn reminder that nudges long autonomous runs to check in.

v2.1.238 lets a plugin marketplace or catalog entry run a headersHelper command that mints short-lived HTTP headers for catalog and archive fetches, so private plugins can sit behind rotating auth without secrets in config. The helper runs only on install or update, after its command is shown, with credential env vars scrubbed for untrusted sources. The release also fixes unbounded memory growth in long sessions, stops output styles drifting to the default voice, and lands a large Remote Control reliability pass. A new readline keybinding flavor makes Ctrl+W delete to the previous whitespace.

Claude Code v2.1.234 continues a session automatically when a claude.ai usage limit resets, toggleable in /config. GitLab repos get a merge request badge in the footer and statusline and MR support in --worktree and the agents view. /goal now clears itself on a dead turn and checks in when background work stalls past CLAUDE_CODE_GOAL_CHECKIN_MINUTES, todo tools are off by default on newer models, and Windows NT-namespace paths are rejected across pre-approval file accesses.

In v2.1.228 the Write tool can overwrite a file that Claude has not read this session, which matches the Edit tool. The change removes one read call and one safety guard. It applies only to newer models through a dated betaAgentToolset20260401 bundle, so older models still read first. The release also recovers frozen sessions, fails fast on stale Google Cloud and AWS logins, and adds CLAUDE_CODE_MESSAGING_TOKEN and NO_HEARTBEAT.

v2.1.229 lets a plugin marketplace resolve from a local command that re-runs each session, so an IDE can keep its own plugin current with no reinstall. CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS staggers same-prefix workflow agents so later ones read a warm prompt cache and send fewer input tokens. The rest is fixes: long responses no longer duplicate or vanish mid-stream, non-string tool inputs stop crashing the session, narrow terminals stop throwing RangeErrors, and Windows handles extended-length paths.

The v2.1.227 changelog lists only fixes: a Fable-tier prompt fix for Max members on expired tokens, a claude-code-action Bash fix, a /tui rewind fix, and a clearer slash-command menu. The build holds three features that no note mentions, all first appearing in this build: a read-only comment-thread-analyst that sorts artifact comments into an edit request or a written reply (CLAUDE_CODE_ARTIFACT_COMMENT_RESPONDER, gate tengu_bracken_sluice); two new env vars (CLAUDE_CODE_MEMORY_API_BASE_URL / _TOKEN) that point Claude Code's memory sync at the Managed Agents Memory Stores REST API under the managed-agents-2026-04-01 beta header; and per-spawn Bash command clamps that limit a subagent's shell to a fixed set of command forms and fail closed. A companion post documents the Memory Stores API in full.

Every route, payload, and status code behind Claude Code's org-memory env vars: the Managed Agents Memory Stores REST API, plus the env vars to point it at your own server.

v2.1.224 lets Team and Enterprise users host Claude Code web, mobile, and desktop sessions on their own machines via `claude self-hosted-runner`, and lets separate sessions message each other across machines with a cross-session SendMessage and ListAgents on macOS and Linux. The messaging socket is local-only and gated, inbound messages from permission-bypassing sessions are held for approval, and the 200-subagent-per-session cap is gone.

Build 0.0.1785875347 ships a plugin attachment upload API (amp.attachments.upload returns a hosted URL for videos, logs, PDFs, and datasets) and fills in the orb runtime: AMP_GIT_PASSWORD and AMP_GITHUB_BOOTSTRAP_TOKEN let a hosted sandbox clone your repo, AMP_DIRECT_TERMINAL attaches your terminal to a running orb, and AMP_ORB_WORKLOAD_SLICE pins it into a systemd cgroup. claude-opus-5 and three Gemini flash-lite tiers join the roster.

v2.1.221 closes a Bash permission-check bypass where zsh could run hidden commands in [[ ]] regex conditionals, and adds a sandbox credential mode that hands sandboxed commands a masked sentinel file while the proxy substitutes the real secret only on egress. A new MCP_PROTOCOL_NEGOTIATION variable lets Claude Code probe servers for a newer protocol revision. The build also carries an unannounced feature: artifact comment threads, where Claude publishes a prototype as a live artifact and reads, replies, and edits in response to human comments, all gated off by default.

v2.1.222 closes the largest gap in worktree isolation: a session in an isolated worktree, and its subagents, could still run destructive git against the main checkout, and isolation now covers file edits and Bash in every session type. It also stops PreToolUse auto-allow hooks bypassing restrictions in background tasks, runs SendMessage through the permission classifier, and fixes the proxy connectivity check, false 'connection closed mid-response' errors, and org model family step-down.

Six stable releases from v0.47.0 to v0.53.1 added no new environment variables, model IDs, or feature flags. The work went into a security cluster (macOS Seatbelt deny-default, read-only ~/.gitconfig in the sandbox, path-traversal and symlink-escape fixes, agent-to-agent RCE prevention), an edit fix that stops the LLM correction pass from mangling JSON and IPYNB files, and a coreTools to tools.core config migration. A new caretaker-agent Cloud Run triage service also landed in tools/, outside the shipped CLI.

Across 15 releases since v1.17.19, OpenCode finished its Desktop v2 migration and added the OPENCODE_SIDECAR_V2 flag, which makes the desktop app launch the new bundled-CLI sidecar instead of the old one. A new subagent_depth config caps how deeply subagents can spawn subagents (default 1), and the desktop client gained right-to-left layout and broader localization. Provider work covered Kimi adaptive thinking, Azure GPT-5.5+ reasoning fixes, and MCP reconnect hardening.

The first Codex CLI release tracked since the 0.144.3 baseline. Conversations become managed threads you can name with /new or /clear, pin, and fork (including temporary forks kept out of listings), on top of 0.145.0's paginated history with resume, search, and memories. Agent Plugins gain manifests, workspace publishing, and marketplaces for Amazon Bedrock and Claude Code, while 0.145.0's /import migrates settings, MCP servers, plugins, and memories from Cursor and Claude Code. The app-server reaches remote Code Mode hosts over WebSocket.

Claude Code v2.1.219 promotes Claude Opus 5 (claude-opus-5) to the default Opus model over claude-opus-4-8, with a native 1M-token context window and $10/$50-per-Mtok pricing. Beyond the model, it clears a run of fixes daily users feel: claude -p no longer drops an answer on a mid-stream error, copy-on-select works again in GNU screen, Vim's left-arrow escapes to the agent view, and screen-reader mode stops rewriting the whole input line.

Claude Code v2.1.218 is mostly cleanup that lands where you feel it. /code-review now runs as a background subagent so it stops flooding your conversation, the left-arrow key can no longer wipe a session with no undo, and a retry loop that re-sent doomed requests after a context overflow is closed. The build also carries four new environment variables the changelog never names: CLAUDE_CODE_AUTO_MODE_CLASSIFIER_QUEUE queues auto mode's per-turn model router, and CLAUDE_CODE_DISABLE_MEMORY_MASS_DELETE_HOLD switches off a default brake that stops one bulk delete from clearing a team's shared memory.

Claude Code v2.1.217 adds twelve environment variables since v2.1.215. The headline pair caps how many subagents run at once (CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS, default 20) and stops subagents spawning their own subagents (CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH, default 1). They are the only additions the changelog names; a FleetView render toggle, an Explore-agent model cap, a transcript garbage-collector, and five codename rollout gates all ship silent inside a 239 MB Bun-compiled executable.

Claude Code v2.1.215 adds five environment variables over v2.1.212: CLAUDE_CODE_USE_ANTHROPIC_GOOGLE_CLOUD (a sixth first-party backend, an Anthropic-operated inference plane on Google Cloud) with its own skip-auth flag, CLAUDE_CODE_NO_MODEL_FALLBACK (pins a session to the exact model you chose, even blocking compaction that would swap models), CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH (60 KB telemetry content cap), and CLAUDE_PID (a shim that stops the agent's own pkill from killing Claude Code). The changelog adds a permission-check hardening pass, a background-session lifecycle cleanup, reasoning effort in the subagent status line, and stops /verify and /code-review from auto-running.

Claude Code v2.1.212 adds eleven environment variables over v2.1.210. The headline additions are two per-session budget caps — CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION and CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION (both default 200) — that halt runaway delegation and search loops, plus a new /bug command wired to a first-party feedback relay (CLAUDE_CODE_SEND_FEEDBACK, gate tengu_juniper_relay) and a Windows-only sandbox gate (CLAUDE_CODE_NANKEEN_KESTREL) that the changelog never names. The official notes confirm the caps and close two quiet security fixes in plan mode and worktree symlinks.

Claude Code v2.1.210 adds 4 new environment variables over v2.1.209. Three wire up a new organization-scoped memory-sync subsystem (CLAUDE_CODE_DISABLE_ORG_MEMORY, CLAUDE_CODE_DISABLE_MEMORY_STREAM_LIST) gated behind two brand-new server gates, tengu_haze_glass and tengu_memory_stream_list, plus remote session-reference propagation (CLAUDE_CODE_SYNC_SESSION_REFS). CLAUDE_CODE_FORCE_WINDOWS_CREDMAN pins credential storage to the Windows Credential Manager. All four ship in this build and none appear in the official changelog.

Daily roll-ups of Amp builds: environment variables, model roster and feature changes, cross-referenced with Amp's public news feed.

A catch-up on Claude Code from v2.1.143 to v2.1.209. One thing stands out of the environment-variable surface: Auto mode stopped being an experiment. Fourteen new CLAUDE_CODE_AUTO_MODE_ variables lay out its safety architecture in unusual detail — an editor model gated by a separate, cheaper classifier, with deletion caps, repo-visibility awareness, external-permission hooks, and structured outcomes. Plus what actually reached users in the last ten releases: Opus 4.8 as the managed default, a safer 'Manual' permission posture, tunable workflows, and a real screen-reader mode.

CLAUDE_CODE_EXPERIMENTAL_OBSERVER_AGENTS is an undocumented Claude Code flag that pairs one agent with a background observer agent. The observer watches the first agent's turns through a read-only digest and can report back through a single ObserverReport tool. First shipped in v2.1.200, it stays silent by default, cannot chain observers, and a server-side gate can still switch it off.

The first tracked snapshot of OpenAI's Codex CLI, read from the open source at tag rust-v0.144.3. The Rust source carries 178 environment-variable names and 40 model-ID strings, with the largest clusters covering sandboxing, a credential-brokering network proxy, cloud task execution, an app and connector marketplace, and a managed-config layer for enterprise fleets. This is the baseline every later Codex CLI post will diff against.

The first tracked release of Google's Gemini CLI. Version 0.47.0 carries 288 environment-variable literals, 39 model-ID strings, and 25 experimental feature flags in its source, and more than half of the variables are telemetry attributes rather than real settings. The configuration knobs that remain cover model routing, multiple auth backends, an extension registry, and a container sandbox; this post is the baseline every later post diffs against.

We analyze every Gemini CLI release to document environment variables, model IDs, and feature changes, grounded in the open-source code at each release tag and the official release notes.

The first tracked snapshot of OpenCode, SST's open-source terminal coding agent, at v1.17.19. The source exposes 120 OPENCODE_ environment variables, 20 of them experimental flags that preview background subagents, parallel execution, workspaces, a native LLM runtime, and LSP-backed code intelligence. A hosted console with orgs, an S3 storage backend, and a control plane sit above the CLI, and this post is the reference every later OpenCode post will diff against.

We analyze every Codex CLI release to document environment variables, model IDs, and feature changes, grounded in the open-source code at each release tag and the official release notes.

We analyze every OpenCode release to document environment variables, config-schema keys, and experimental flags, grounded in the open-source code at each release tag and the official release notes.

27 new env vars, 2 removed across 15 releases. /goal ships a real completion tracker for agentic work. Fast mode silently upgraded to Opus 4.7 at no extra cost. CLAUDE_CODE_MAX_TURNS hard-caps turns; CLAUDE_CODE_SUPERVISED tells Claude a human is watching — the agentic safety pair that was missing. Agent View replaces Agents Fleet naming. Hook args exec form for shell-less subprocess spawning. MCP servers survive /clear. Plugin marketplace shows context cost before install. CLAUDE_CODE_STOP_HOOK_BLOCK_CAP ends the Ctrl+C hang. Three new BG session vars document the failure modes.

2 new env vars. CLAUDE_CODE_PROACTIVE disables the existing UI prompt on /resume and --resume that offers a /compact-equivalent before resuming long conversations — useful for non-interactive / scripted resume flows (the prompt itself has shipped for several versions; only the disable flag is new). CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT strips the system prompt for background subagents, cutting the largest fixed per-turn token cost across parallel agent trees. Sonnet 4.6 takes over as the embedded default for Claude Code Remote. (Original framing of CLAUDE_CODE_PROACTIVE was wrong — see correction note on the post.)

v2.1.124–v2.1.125 were internal builds. v2.1.126 ships claude project purge for surgical cleanup of transcripts and config, a gateway-aware /model picker that queries your actual /v1/models endpoint, --dangerously-skip-permissions covering the last CI holdouts (.claude/, .git/, .vscode/), OAuth paste-in for WSL2/SSH/containers, and a security fix where managed sandbox restrictions silently dropped when the org policy had no sandbox block. Plus 7 bug fixes: stream timeout after sleep, image paste breaking sessions, Windows clipboard exposure, CJK garbled on Windows, Ctrl+L erasing input, deferred tools missing in fork subagents. CLAUDE_CODE_COLD_COMPACT enables safe session resumption after sleep or network drop.

30+ new env vars across five releases. The background daemon graduates from a single flag to a fully specified IPC contract (rendezvous socket, PTY recording, bridge reconnect protocol). Sessions become first-class addressable objects. An agent fleet layer takes shape with selection, auto-relaunch, and a kill switch. CoWork memory is the first inter-agent coordination primitive — a shared RAG layer with policy-governed writes. Daily wins: /config persists, alwaysLoad for MCP, PostToolUse hooks rewrite any tool output, claude ultrareview as a CLI subcommand, default effort raised to high for Pro/Max, three memory leaks fixed.

10 new env vars across four releases. Forked subagents enabled on external builds via CLAUDE_CODE_FORK_SUBAGENT=1 — Agent Teams multi-agent architecture leaves internal-only status. Subscription and rate-limit tier vars describe a runtime moving toward per-plan capability differentiation. Named proxy vars simplify corporate proxy setup. Bridge reconnection handshake and retry watchdog close the remaining silent-hang surface. Also covers Anthropic's April 23 postmortem on the March–April quality regression and usage limit reset.

Seven-release stretch from v2.1.107. Two headline shifts: v2.1.113's CLI now spawns a native per-platform binary instead of bundled JS (tighter supply chain, faster cold start). And v2.1.111's /fewer-permission-prompts skill scans transcripts to propose a prioritized settings allowlist — the first feature that actually cuts Claude Code's biggest daily friction. Plus fullscreen TUI, /ultrareview cloud PR review, Opus 4.7 xhigh effort, 1-hour prompt caching, proxy auth helper protocol for Kerberos/SSO (5 vars), Agent Teams scaffolding, experimental Advisor Tool. 20 new env vars, 1 new model ID, ~80 bug fixes.

6 new env vars, 0 removed. Multi-worktree orchestration: base ref overrides (CLAUDE_CODE_BASE_REFS), worktree storage paths (CLAUDE_CODE_REPO_CHECKOUTS), session chaining (CLAUDE_CODE_RESUME_FROM_SESSION). Away mode summaries for async agent supervision. Byte-level stream watchdog catches stalled API connections below SSE layer. PreCompact hooks, background plugin monitors, worktree reuse via path parameter.

4 new env vars, 0 removed. Limits reckoning: prompt cache TTL now user-configurable, rate limits vs plan limits get separate errors, startup warning if caching disabled, DISABLE_TELEMETRY → 5-min TTL bug fixed. REPL experimentation returns with CLAUDE_CODE_REPL and CLAUDE_REPL_VARIANT for A/B testing stateful execution. Skill discovery via Skill tool. /undo alias, /recap away summary, lazy-loaded grammars, 15+ bug fixes.

2 new env vars, 0 removed. Enterprise TLS certificate store control (CLAUDE_CODE_CERT_STORE) — OS CA store trusted by default, bundled reverts to Node.js CAs. SDK OAuth refresh coordination (CLAUDE_CODE_SDK_HAS_OAUTH_REFRESH) prevents double-refresh races in IDE extensions. /team-onboarding command. /ultraplan auto-provisioning. Command injection fix in LSP binary detection. Virtual scroller memory leak fix. Hardcoded 5-minute timeout removed.

3 new env vars, 3 removed. Context token ceiling (CLAUDE_CODE_MAX_CONTEXT_TOKENS) with DISABLE_COMPACT interop. Perforce VCS mode — fail on read-only files with p4 edit hint. Script capability caps paired with PID namespace sandboxing. Google Vertex AI setup wizard. Focus view toggle. Monitor tool for background scripts. 30+ Bash permission security fixes. MCP memory leak fix. REPL experiment removed.

0 new env vars, 0 removed. Bedrock auth hotfix — fixed 403 "Authorization header is missing" regression from v2.1.94 when using AWS_BEARER_TOKEN_BEDROCK or CLAUDE_CODE_SKIP_BEDROCK_AUTH. Tracked surface identical to v2.1.94. Mantle auth layer still stabilizing. v2.1.95 skipped.

5 new env vars, 0 removed. Mantle auth proxy for Amazon Bedrock (CLAUDE_CODE_USE_MANTLE) with skip flag. Sandbox state flag (CLAUDE_CODE_SANDBOXED). MCP server environment variable allowlist replaces all-or-nothing env inheritance. Team onboarding flows. --resume across worktrees. CJK text corruption fix. Write tool diff 60% faster.

5 new env vars, 0 removed. Proxy simulation without infrastructure (CLAUDE_CODE_SIMULATE_PROXY_USAGE). Fast mode org check bypass. Remote Control session name prefix for CI/CD observability. CLAUDE_CODE_EXECPATH for symlinked binary overrides. CLAUDE_BASE shorthand. Interactive Bedrock setup wizard. Per-model cost breakdowns. Policy-enforced remote settings refresh.

1 new env var, 2 removed. MCP 500K character tool result persistence override. Plugin bin/ executable support — native binaries in plugins. disableSkillShellExecution setting locks down skill shell access. Multi-line deep links. Edit tool token optimization with shorter old_string anchors. CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION replaces typo predecessor.

3 new env vars, 1 new command. Anthropic AWS as fifth API provider (CLAUDE_CODE_USE_ANTHROPIC_AWS) with IAM ambient auth skip. Offline-resilient marketplace plugin caching. /powerup interactive lessons. Prompt-cache regression fix since v2.1.69. Quadratic SSE transport and transcript write fixes. PowerShell security hardening. Auto mode now respects explicit user constraints.

4 new env vars, 2 removed. Agent cost steering for subagent delegation. Session resume thresholds by time and token count. Claude API skill disable flag. Deferred tool execution for headless sessions. StructuredOutput schema cache ~50% failure rate fixed. Autocompact thrash loop detection. Edit tool no longer requires prior Read. MCP non-blocking connections. Command injection check now mandatory.

2 new env vars, 2 removed. Net count flat at 202. OAuth console override completes the auth triptych. Trusted device token for persistent sessions. MCP server env vars removed after 2 versions. Cowork Dispatch delivery fix plus model selection. Boris Cherny's 15 hidden features thread. --bare default flip coming.

7 new env vars, 3 removed. Terminal UX overhaul: CLAUDE_CODE_NO_FLICKER for flicker-free rendering, granular mouse control (CLAUDE_CODE_DISABLE_MOUSE, CLAUDE_CODE_DISABLE_MOUSE_CLICKS), tunable scroll speed. REPL mode emergence with CLAUDE_CODE_REPL and CLAUDE_REPL_MODE for stateless scripting workflows. Policy skill suppression. v2.1.87 Cowork Dispatch hotfix. Prompt cache miss fix, structured output schema bug, memory leaks for 1 GiB+ files.

6 new env vars, 1 removed. MCP server multiplexing with CLAUDE_CODE_MCP_SERVER_NAME and CLAUDE_CODE_MCP_SERVER_URL. OAuth endpoint overrides for enterprise identity providers. PowerShell parse timeout. Executable shrank 7.15 MB by replacing WASM yoga-layout (yes, flexbox in your terminal) with pure TypeScript. Conditional hook if fields. Scheduled task timestamps. Memory leak patches.

2 new env vars. CLAUDE_TRUSTED_DEVICE_TOKEN for enterprise SSO and CI/CD device trust. CLAUDE_LOCAL_OAUTH_CONSOLE_BASE completes the OAuth endpoint decomposition — every auth touchpoint now redirectable to corporate infrastructure. 20+ bug fixes: --resume tool_use ID errors, /feedback OOM, masked input token leaks, MCP startup stalls.

6 new env vars. Idle session management with configurable thresholds, stream watchdog timeout. Bedrock/Vertex/Foundry capability override env vars. TaskCreated and WorktreeCreate HTTP hooks. Deep links open in preferred terminal. PowerShell tool preview. MCP 2KB cap and deduplication. Stats screenshot (Ctrl+S in /stats). Windows drive root protection.

8 new env vars, 40+ bug fixes, v2.1.82 skipped. managed-settings.d/ drop-in directories for enterprise policy, CLAUDE_CODE_SUBPROCESS_ENV_SCRUB for credential scrubbing, transcript search, plugin keychain storage, CwdChanged/FileChanged hook events. macOS exit hang fixed, startup latency cuts, MCP policy bypass closed.

The --bare flag strips Claude Code to its core for scripted -p calls — no hooks, LSP, plugins, or auto-memory. Concurrent OAuth fix ends the multi-session re-auth nightmare (#24317). Channels permission relay, /btw pasted text fix, ! bash mode discoverability. Plan mode hides clear context. 17 bug fixes.

Channels research preview turns Claude Code into a persistent agent reachable from Telegram and Discord. Simplified plugin install, inline settings declarations, 80MB memory reduction for large repos. Fine-grained streaming fix retires the CLAUDE_CODE_ENABLE_FINE_GRAINED_TOOL_STREAMING=0 workaround. 7 bug fixes.

3 new env vars, 2 removed: subprocess environment scrubbing for security, extra API metadata for enterprise routing, refactored init path saving 18MB startup memory. Console auth for API billing, non-streaming fallback timeout, 10+ bug fixes.

1 new env var and 1 renamed slash command: cowork memory extra guidelines for multi-agent teammate sessions, /branch (renamed from /fork), and /copy N index. Largest stabilization release tracked — 25+ bug fixes spanning terminal rendering, tmux, teammate sessions, and memory management.

2 new env vars: plugin persistent state directory and custom model picker entries. StopFailure hook event, plugin agent frontmatter expansion (effort, maxTurns, disallowedTools), line-by-line streaming. Security hardening: sandbox bypass fixes, MCP deny rule enforcement, protected directory prompts.

3 new env vars and 1 new slash command: account tagged IDs for multi-tenant billing, fast-mode network error resilience, workspace host path mappings for containers, and /effort for interactive effort level control. Headline feature: MCP elicitation support for bidirectional server communication.

2 new env vars and 1 new slash command: auto-compact window threshold for 1M context windows, plugin option prefix for CI/CD configuration, and /color for session-level prompt-bar customization.

Small release with 2 new env vars and 1 removal: frame timing diagnostics, session-end hooks timeout, and the two-stage classifier flag graduated to default.

Maintenance release with no env var changes. First version with the full slash command inventory tracked: 22 built-in commands form the baseline.

Minor release with 6 new env vars: brief mode controls, debug log level, cron disable, virtual scroll toggle, and two-stage classifier.

Minor release with 3 new env vars: CCR v2 bridge protocol toggle, force-enable effort level, and delta MCP instruction updates. Plus one removal.

Claude Code v2.1.69, 50 versions after the original v2.1.19 post. The count nearly doubled: 165 env vars, 16 model IDs, 41 feature gates. New plugin system, mTLS support, background tasks, context compaction controls, and multi-provider auth.

Claude Code 2.1.19 ships 83 undocumented environment variables. None are in the official docs. Highlights include Agent Swarms, Plan v2 agent controls, debug/profiling flags, and ways to disable features like CLAUDE.md loading.