NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules
Go modules, #1299 by repository stars
Latest stable
v0.83.2
Last release 7 days ago
15 Nov 2026
Ships on a steady schedule
a new release about every 8 days
Nearly every release is documented
notes for 59 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
6 months old
289 releases · first in 2026
Drag across the ruler to compare, or tap a release to open it.
Managed SSH hosts: prepare public enrollment descriptors and install per-host expiring credentials, with admin-only revocation, reissue and exact-cust
Your coding agent can read these notes before it upgrades. Set up the MCP server →
One column per month.
Nothing published for this version
Nothing published for this version
Azure Sandbox: select every resource tier from XS (0.25 vCPU / 0.5 GiB) to 3XL (16 vCPU / 32 GiB), including 5–320 GiB tier disk capacities, with --ty
--type names or tier aliases while preserving implicit defaults, safe lease replay and snapshot inheritance. https://github.com/openclaw/crabbox/pull/3143 Thanks @galiniliev.Nothing published for this version
Nothing published for this version
Fuel evidence: add an optional source-built Wasmtime helper and native qualification through existing SSH artifacts, with waited-child CPU diagnostics
Nothing published for this version
Nothing published for this version
Keep long Linux runs stable through scheduled update windows. Provisioned Debian/Ubuntu guests no longer self-apply scheduled APT updates that can int
crabbox run | head finishes one-shot cleanup before returning exit 141 on Unix. PR 3041, PR 3040. Thanks @lazmo88 and @coygeek.crabbox run now exits 141 after cleanup when its output consumer closes early; account for that status in pipelines using pipefail. PR 3040. Thanks @coygeek.external, and reject stale summaries in docs CI. PR 3034. Thanks @coygeek.keep=true does not exempt expired machines and the running or provisioning grace is one hour. PR 2999. Thanks @KrasimirKralev.actions unregister without requiring the VM to remain reachable. PR 3028. Thanks @vincentkoc.run and warmup. PR 3044. Thanks @coygeek.nomodule boot argument so module-capable guests cannot load kernel modules and init no longer receives nomodules. PR 3032. Thanks @coygeek.Nothing published for this version
Nothing published for this version
Actions on Linux: keep hydration metadata and environment handoff within bounded control commands while preserving workspace ownership. PR 3023. Thank
Nothing published for this version
Nothing published for this version
Checkpoint and inspect native Hyper-V Windows guests. Save, restore, and fork checkpoints with fresh identities, and capture interactive desktops, scr
wb CLI. PR 3004. Thanks @vincentkoc.crabbox pause before planned maintenance to preserve memory. crabbox resume rejects powered-off VMs with exit 4; release them with crabbox stop. Existing leases retain their original setting. PR 2991. Thanks @paulcam206 and @RomneyDa.asciiBox.noEnv or --ascii-box-no-env, retaining creation policy across fixed leases and blocking incompatible execution reuse. PR 3010. Thanks @vincentkoc.--type or Crabbox classes for ordinary and fixed-ID leases, preserving existing default-sized lease recovery. PR 3003. Thanks @vincentkoc.wb CLI. PR 3004. Thanks @vincentkoc.crabbox cp over SFTP, private staging, size and SHA-256 checks, and atomic destination replacement under the exact lease claim. PR 2988. Thanks @paulcam206 and @Copilot.--desktop, probe-only --browser, and screenshot, video, and desktop-proof support for native Windows guests, capturing through the lease user's interactive session. PR 2990. Thanks @paulcam206 and @RomneyDa.--debug while preserving full-manifest fallback and keeping raw probe output private. PR 3016. Thanks @vincentkoc.AutomaticStopAction=TurnOff for new leases so running guests do not reserve memory-sized state files on the VM drive. PR 2991. Thanks @paulcam206 and @RomneyDa.BOAT_API_KEY is present. PR 3005. Thanks @vincentkoc.run has no --workdir flag. PR 2982. Thanks @KrasimirKralev.Share existing Linux and Windows machines through the coordinator. Managed hosts use outbound relays, isolated borrower execution, rotating pool acces
static.noControlMaster, CRABBOX_STATIC_NO_CONTROL_MASTER, and --static-no-control-master to opt out of SSH multiplexing, preserving explicit overrides in recovery commands. PR 2855. Thanks @shunkakinoki.still seeding remote Git metadata after ... progress lines while a remote Git seed runs, so long seeds of large repositories no longer look stalled. PR 2974.--concurrency or shard.maxConcurrency, preserving all-shards-at-once defaults and cancellation cleanup. PR 2935. Thanks @zozo123.exit=cancelled, exit=deadline, or exit=signal:<name> instead of exit=unknown. PR 2973.static.id / CRABBOX_STATIC_ID shares the host safely, and warns that --reclaim takes over the other repository's lease and workdir. PR 2972.runsc or Kata), validated strictly and recorded in lease ownership; the default runtime is unchanged. PR 2950. Thanks @coygeek.crabbox, or a generic ssh.user / CRABBOX_SSH_USER) is honored instead of being replaced by the host USER; the host-user fallback still applies when no guest user is configured. PR 2958. Thanks @rustanacexd.XDG_STATE_HOME by comparing handle-based file identities while retaining managed-state path replacement checks. PR 2963. Thanks @paulcam206.Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Refresh managed developer tools with verified security updates. Bun 1.4.4, uv 0.13.5, TruffleHog 3.99.4, Tailscale 1.104.2, and Git for Windows 2.56.0
azure.tenantId / azure.subscriptionId keys and their environment variables instead of nonexistent azure.tenant / azure.subscription keys. PR 2943. Thanks @KrasimirKralev.Nothing published for this version
Nothing published for this version
Nothing published for this version
Keep Windows values intact and prevent PowerShell injection. Typographic single quotes stay inside string literals across commands, forwarded environm
serve under self-hosted-virtualization, generate the category index from metadata, and check README category membership. PR 2936. Thanks @aitrawog.Nothing published for this version
Patch certificate validation across Go builds. Source, release binaries, container runners, and developer-image toolchains now select Go 1.27.3 to fix
Patch certificate validation across Go builds. Source, release binaries, container runners, and developer-image toolchains now select Go 1.27.3 to fix GO-2026-6809, retaining the Go 1.26 minimum.
Local Actions hydration: hydrate detached source mirrors from a verified published ancestor with only a local base ref, preserving selected-file exclusions during cold sync. PR 2916. Thanks @shakkernerd.
Local Actions hydration: prepare cold unpublished detached sources from a verified published ancestor while preserving selected files, privacy filters, and repository identity checks. PR 2914. Thanks @shakkernerd.
Patch certificate validation across Go builds. Source, release binaries, container runners, and developer-image toolchains now select Go 1.27.3 to fix GO-2026-6809, retaining the Go 1.26 minimum.
Local Actions hydration: hydrate detached source mirrors from a verified published ancestor with only a local base ref, preserving selected-file exclusions during cold sync. PR 2916. Thanks @shakkernerd.
Local Actions hydration: prepare cold unpublished detached sources from a verified published ancestor while preserving selected files, privacy filters, and repository identity checks. PR 2914. Thanks @shakkernerd.
machine:files scope. PR 2915. Thanks @NickyHeC.Nothing published for this version
Nothing published for this version
Release owned resources with stronger replacement checks. Tart validates recorded VM ownership, AWS matches immutable key-pair IDs, Azure checks snaps
uniqueId, or with a mismatched identity, become detached instead of deleting by name. Reconcile those snapshots manually; --local-only removes only the local record. PR 2905. Thanks @coygeek.CI: make Blacksmith's buffered artifact-timeout regression deterministic under scheduler pressure without changing production timeouts or workload results. PR 2909.
Config: protect persisted Windows credentials with a verified current-user-only DACL before writing, without inheriting parent read grants. PR 2901. Thanks @coygeek.
Freestyle: create and resume VMs through the current Freestyle API contract, validating readiness for stopped or paused creation responses. PR 2897. Thanks @coygeek.
Coordinator: revoke non-admin control sessions after shared-token rotation or GitHub authorization withdrawal, including restored sessions, event delivery, and queued heartbeats. PR 2907. Thanks @coygeek.
Freestyle: report commands whose exec response has no exit status (timeouts) as failures instead of success, keeping output and failure retention. PR 2896. Thanks @coygeek.
Tests: preserve literal SSH username and transport coverage across OpenSSH parser versions, including clients that reject combined backslash/percent usernames. PR 2904. Thanks @coygeek.
Windows Sandbox: stop only captured run-owned process identities during cleanup, preserve replacement sessions, and report unconfirmed ownership. PR 2906. Thanks @coygeek.
AWS: release deletes the lease key pair only by its recorded immutable key-pair ID and skips replacement keys registered under the reused name, explaining every skip. PR 2890. Thanks @goutamadwant and @coygeek.
Azure: bind direct checkpoint deletion to the captured snapshot identity and detach stale or legacy records for manual cleanup. PR 2905. Thanks @coygeek.
Login: preserve configured broker provenance during browser authentication so repository-selected destinations cannot use inherited coordinator credentials. PR 2903. Thanks @coygeek.
Tart: enforce recorded VM ownership and current claims during explicit release, preserve retry state on failure, and verify storage absence before pruning missing leases. PR 2902. Thanks @coygeek.
Sync: preserve Darwin symlink permissions in local Git seed and overlay snapshots without changing symlink targets or weakening source validation. PR 2900. Thanks @vincentkoc.
Sync: retain redacted attempt, stage, and failed-check diagnostics when local Git-seed snapshot preparation exhausts its retries, without changing acceptance or cleanup. PR 2899. Thanks @vincentkoc.
CI: move macOS checks and release verification to macOS 27, preserving macOS 13 deployment gates and Intel verification through Rosetta. PR 2881.
Docs: clarify command-specific --provider selection, doctor readiness failures, and providers describe <name> for catalog inspection. PR 2873. Thanks @aitrawog.
Static SSH: derive the macOS work root from the final --static-user flag instead of the default or config user, while explicit work roots still win and the generic SSH user remains the fallback. PR 2871. Thanks @holgergruenhagen.
Sync: stage rsync receivers through an explicit POSIX shell for Fish-login SSH hosts and identify receiver setup failures without weakening workspace ownership. PR 2874. Thanks @shunkakinoki.
Run: retry workspace-owner RENEWED replies with transport errors within the existing same-token retry budget and confirmed deadline; only an error-free reply confirms ownership. PR 2872. Thanks @shunkakinoki.
Coordinator: read run-log previews from the latest chunks without scanning the full retained log. PR 2869. Thanks @vincentkoc.
CI: stop aggregate checks when a run is cancelled while preserving failed dependency results, and check coverage producers before toolchain setup. PR 2868. Thanks @vincentkoc.
Security: update the Incus client and HTTP/2 dependencies to address reachable advisories, and use Go 1.27.2 for builds while retaining the Go 1.26 module minimum. PR 2870.
Coordinator deployment: reject lease-policy overrides the runtime would ignore, while preserving acknowledged repair of invalid live policy values. PR 2867. Thanks @vincentkoc.
Coordinator: reconcile retained AWS records with filtered batches of at most 200 unique instance IDs, avoiding missing-instance fallback storms and isolating transient chunk failures while preserving successful observations and bounded backoff. PR 2866.
Sync: keep finalization fetches blob-filtered when completing shallow Git history, and explain live-child owner release refusals with safe lease recovery guidance. PR 2854.
Tests: use small controller fixtures in archive tests while preserving companion and manifest validation. PR 2876. Thanks @vincentkoc.
CI: use keyless OIDC intake for review dispatch. PR 2892.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Put busy-but-usable machines back in the pool. Heavily loaded Macs now report real availability instead of staying unknown and never being offered for
unknown and never being offered for new leases. PR 2845.checkpoint create --wait=false now establishes every backing snapshot before returning, so subsequent work cannot race an unstarted capture. PR 2834.sa-west-1 root CAs (114 Amazon RDS certificates, all self-signed Amazon roots). PR 2853.captureEstablished attestations while preserving other providers' success semantics, and reduce native availability and AWS public-IP polling slack. PR 2834.Donate spare capacity without slowing down your own work. crabbox serve offers new leases only when host CPU, memory, user activity, and power policie
crabbox serve offers new leases only when host CPU, memory, user activity, and power policies allow; running leases continue, and the CLI and portal explain why a machine is busy. PR 2838.cleanup phase instead of claiming the machine has stopped. PR 2819. Thanks @vincentkoc.lease.released must use the cleanup phase and leaseStopped=false to distinguish an accepted release request from confirmed machine cleanup. PR 2819. Thanks @vincentkoc.--availability always only when a donor should accept new leases regardless of host activity; laptop defaults also require ten minutes of user idle and AC power. PR 2838.cleanup phase. PR 2819. Thanks @vincentkoc.Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Production lease policy gives owner github:58493 a 12-hour maximum lease lifetime for long serial proof runs; the 30-minute default and idle timeout s
github:58493 a 12-hour maximum lease lifetime for long serial proof runs; the 30-minute default and idle timeout still apply.stop --wait=false for coordinator-owned queued cleanup while preserving pending local recovery state; ordinary stop still waits for confirmed deletion.crabbox stop --wait=false hands confirmed deletion to the coordinator and keeps local recovery state until cleanup completes; plain stop still waits. PR 2809.stop --wait=false for coordinator-owned queued cleanup while preserving pending local recovery state; ordinary stop still waits for confirmed deletion. PR 2809.Nothing published for this version
CI: run CLI coverage in three isolated shards, retaining atomic whole-suite coverage and the 90% core gate; report the slowest top-level tests without
crabbox list streams and shows live leases by default; --retained includes kept terminal leases. Add --jsonl streaming and uncapped --all lease lookups while preserving the buffered --json array.keep=true leases and reduces running/provisioning expiry grace from 12 hours to 1 hour; unowned machines remain untouched.blacksmith.waitTimeout, CRABBOX_BLACKSMITH_WAIT_TIMEOUT, or --blacksmith-wait-timeout, preserving the native 15-minute default when unset.crabbox admin retained-reconcile status.directLeases: deny and trusted provider exceptions to require coordinator allocation.crabbox extend requests bounded lifetime extensions; running commands can auto-extend when policy and provider lifetime capabilities allow.t3.small with class tiny and On-Demand capacity so inherited defaults cannot launch a larger instance. PR 2814.CRABBOX_SERVE_RELAY_DEBUG=1. PR 2811.reservedUSD in usage reports and expose enforced cost as committedUSD, with compatible CLI output and conservative reservations for retained legacy VMs and unresolved provisioning. PR 2801.crabbox extend and in-run auto-extension are budget-checked. PR 2804.keep=true machines are no longer exempt from cleanup, and directLeases: deny requires coordinator allocation on a machine. PR 2806.crabbox list answers in seconds. Listings stream live leases first and show only live leases by default; --retained adds kept history, --jsonl streams machine-readable rows, and --all keeps orphaned provider inventory visible without a 1,000-row cap. PR 2802.crabbox admin retained-reconcile. PR 2803.crabbox run auto-extends while its command executes; request a longer --ttl or use crabbox extend for interactive work. Daytona and other creation-fixed providers cannot be extended after creation. PR 2804.crabbox list and crabbox list --json now return only active and provisioning leases; add --retained where scripts relied on kept terminal records. PR 2802.crabbox cleanup for direct providers now deletes expired machines labeled keep=true and gives running or provisioning machines a 1-hour grace instead of 12 hours. PR 2806.lease_extension_budget_exceeded. PR 2804.crabbox list streams and shows live leases by default; --retained includes kept terminal leases. Add --jsonl streaming and uncapped --all lease lookups while preserving the buffered --json array. PR 2802.keep=true leases and reduces running/provisioning expiry grace from 12 hours to 1 hour; unowned machines remain untouched. PR 2806.crabbox extend requests bounded lifetime extensions; running commands can auto-extend when policy and provider lifetime capabilities allow. PR 2804.directLeases: deny and trusted provider exceptions to require coordinator allocation. PR 2806.crabbox admin retained-reconcile status. PR 2803.blacksmith.waitTimeout, CRABBOX_BLACKSMITH_WAIT_TIMEOUT, or --blacksmith-wait-timeout, preserving the native 15-minute default when unset. PR 2805. Thanks @vincentkoc.CRABBOX_SERVE_RELAY_DEBUG=1. PR 2811.t3.small with class tiny and On-Demand capacity so inherited defaults cannot launch a larger instance. PR 2814. Thanks @vincentkoc.reservedUSD in usage reports and expose enforced cost as committedUSD, with compatible CLI output and conservative reservations for retained legacy VMs and unresolved provisioning. PR 2801.Nothing published for this version
Turn spare machines into private container capacity. crabbox serve offers owner-only leases through one authenticated SSH front door, with pinned end-
crabbox serve offers owner-only leases through one authenticated SSH front door, with pinned end-to-end container SSH, warm package caches, and opt-in capacity fallback. PR 2768, PR 2770, PR 2789.jj workspaces now sync their exact @ tree, including working-copy edits and octopus merges, to ordinary SSH leases. PR 2780, Issue 1081. Thanks @atimmer.crabbox run should allow up to 30 seconds for run admission when the coordinator is unavailable, increased from 10 seconds; shorter caller deadlines still win. PR 2777.crabbox list --all now paginates active-lease lookups, so results can include active leases beyond the former single-page cap. Older coordinators retain the legacy fallback with a warning when its limit is reached. PR 2788.@ tree, including octopus merges and working-copy edits, with explicit origin-bookmark seeding and fail-closed revision checks. PR 2780, Issue 1081. Thanks @atimmer.crabbox serve on trusted machines to donate spare CPU as owner-only container leases (--provider serve), with outbound coordinator polling, local deadline enforcement, a single authenticated SSH front door (--ssh-port) with pinned end-to-end container SSH, warm package caches, per-user service install, and opt-in capacity fallback. Coordinators enable it per owner with CRABBOX_SERVE_ALLOWED_OWNERS (off by default). PR 2768, PR 2770, PR 2789.serve_rate_limited heartbeat errors by pacing each heartbeat at least --poll-interval after the previous response. PR 2770.Nothing published for this version
Nothing published for this version
Nothing published for this version
Recover GitHub Codespaces leases safely after interrupted requests. Caller-supplied lease IDs now bind recovery to the exact account and creation inte
--exe-dev-from to copy a ready workspace into a separately owned lease with fresh ownership tags and a persistent work root. PR 2757, PR 2754. Thanks @salmonumbrella.capacity=unknown quota_limit=sufficient usage=unchecked region=...: a sufficient quota no longer implies available regional capacity. PR 2753. Thanks @vincentkoc.--exe-dev-from, fresh ownership tags, and persistent work roots while preserving existing claim paths. PR 2757, PR 2754. Thanks @salmonumbrella.Nothing published for this version
Nothing published for this version
Nothing published for this version
Start AWS Windows leases on Server 2025. The promoted eu-west-1 developer image and stock fallback now use Windows Server 2025; an explicit OS selecto
crabbox hold for salvage. PR 2738, PR 2733. Thanks @galiniliev.--os windows-server:2022 to keep Server 2022. PR 2748, PR 2734, Issue 2717. Thanks @saftall.crabbox stop or crabbox hold explicitly, even for an expired lease; retain local claims for recovery. Holds do not expire automatically and need explicit finalization after salvage and disposal. PR 2733. Thanks @galiniliev.CRABBOX_IMAGE_CAPACITY_WAIT (default 45 minutes per acquisition) for fleet, owner, or org active-lease capacity, preserving cleanup and excluding rejected attempts from measured timings. PR 2742.--target windows while incompatible final selections still fail. PR 2737.