Open
Conversation
c1e1b5e to
64fa1d6
Compare
fe0733f to
672422c
Compare
7f3d79f to
b49c7c7
Compare
098337e to
82a7b5a
Compare
bd03337 to
2118436
Compare
2118436 to
98adc4d
Compare
b6fbefc to
eaf86e7
Compare
18bcb2c to
e001f7f
Compare
a3deabf to
2c32896
Compare
2c32896 to
eb5a672
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR contains the following updates:
v2024.7.1→v2026.3.8Release Notes
jdx/mise (jdx/mise)
v2026.3.8: : Wrapper recursion fix and lockfile provenance correctionCompare Source
A small patch release that fixes infinite recursion in
mise execwhen wrapper scripts and shims coexist in PATH, and corrects lockfile provenance detection for aqua tools with opts-only cosign configurations.Fixed
mise execno longer infinite-loops when wrapper scripts and shims are both in PATH -- In devcontainer setups (and similar environments), a wrapper script like.devcontainer/bin/gitleaksthat callsmise x -- gitleakswould resolve back to itself instead of the real binary when the wrapper directory appeared before the shims directory in PATH. This caused infinite recursion until the environment exceeded ARG_MAX, producing confusing errors. The fix reorders the internal lookup PATH so that mise-managed tool bin directories are checked before system PATH entries, ensuring the real binary is always found first. The child process still inherits the full unmodified PATH. #8560 by @jdxLockfile no longer records unverifiable cosign provenance for opts-only aqua tools -- Tools like
yamlfmt,trufflehog, andtflintconfigure cosign in the aqua registry with only CLI opts (e.g.--certificate-identity) but nokeyorbundleconfig.mise lockwas recordingprovenance = "cosign"for these tools, butmise installcan only verify cosign natively via key-based or bundle-based flows, causing "Lockfile requires cosign provenance ... but no verification was used" errors on install. Now cosign provenance is only recorded when the tool has a key or bundle config that can actually be verified. #8559 by @jdxAdded
turboadded to the registry -- Turborepo, the high-performance build system for JavaScript and TypeScript codebases, can now be installed viamise use turbo. #8553 by @igasworkmuxadded to the registry -- workmux, a tool for git worktrees + tmux windows for zero-friction parallel development, can now be installed viamise use workmux. #8555 by @ifraixedesNew Contributors
Full Changelog: jdx/mise@v2026.3.7...v2026.3.8
v2026.3.7: : Cleaner conda PATH handlingCompare Source
A small patch release that fixes conda packages polluting PATH with dozens of transitive dependency binaries.
Fixed
conda:postgresqlwould previously add every binary from every transitive dependency to PATH. For example,conda:postgresqlexposed 106 binaries including ncurses utilities (clear,reset,tput,tabs), openldap commands (ldapadd), and krb5 tools (kinit) -- all of which could shadow standard system commands. Now only binaries belonging to the main requested package are placed on PATH (e.g.,psql,pg_dump,createdb,initdb,pg_ctl,postgres, etc.). Dependency binaries remain installed and available to packages that need them internally, but are no longer visible on PATH. No user configuration is needed -- this is automatic for all new conda installs, and existing installs gracefully fall back to the previous behavior. #8543 by @simonepriNew Contributors
Full Changelog: jdx/mise@v2026.3.6...v2026.3.7
v2026.3.6: : Per-environment lockfiles, Windows fixes, and fork bomb preventionCompare Source
This release redesigns environment-specific lockfiles for better CI isolation, fixes a critical fork bomb caused by shim recursion in
go:backend tools, and addresses several Windows-specific issues including locked.exeshims andenv._.sourcesupport.Highlights
envtag system, somise.test.tomlnow generatesmise.test.lockinstead of tagging entries in a sharedmise.lock. This means CI caches are no longer invalidated by dev-only tool changes.go:backend tools are configured alongside an uninstalled Go version in shims mode..exeshim handling gracefully renames locked shim executables instead of failing with "Access is denied" during reshim.Changed
Lockfiles are now per-environment -- Environment-specific configs now get their own lockfiles (
mise.test.toml->mise.test.lock,mise.local.toml->mise.local.lock) instead of usingenv = ["test"]tags in a singlemise.lock. This improves CI cache isolation -- environments that don't setMISE_ENVonly depend onmise.lock, so dev tool version bumps won't invalidate CI caches. Old lockfiles withenvfields are silently accepted and migrated on the nextmise lock. #8523 by @jdxmise.tomlmise.lockmise.test.tomlmise.test.lockmise.local.tomlmise.local.locktouch_outputsremoved from prepare providers -- Thetouch_outputsconfiguration option has been removed from prepare providers. Freshness checking now uses blake3 content hashing exclusively, so touching output modification times is no longer necessary. #8535 by @jdxFixed
Fork bomb when using
go:backend tools in shims mode -- When ago:backend tool (e.g.,go:github.com/pulumi/upgrade-provider) was configured alongside a Go version that wasn't installed, and the version cache was cleared, mise could enter infinite shim recursion -- the Go shim would callmise exec, which would resolve thego:backend, which would callgo listvia the shim, and so on. The fix strips mise's shims directory from the PATH independency_env, ensuring dependency tools either resolve to a real installed binary or fail cleanly. This applies to all backends that use dependency environments (go,npm,gem,dotnet,spm,elixir). #8475 by @poseLocked
.exeshims on Windows during reshim --mise reshimwould fail with "Access is denied" on Windows when.exeshims were locked by running processes. The fix removes shims individually (instead of wiping the entire directory) and uses a rename-to-.oldfallback for locked files, which Windows allows even when the file is in use. The.oldfiles are cleaned up on the next reshim. #8517 by @davireisenv._.sourcenow works on Windows --env._.sourcewould fail on Windows because it searched forbashwithout the.exeextension. The Windows API executable search now correctly looks forbash.exe. #8520 by @pjebyGitHub
@latestversion resolution -- Thegithub:backend would fail with a 404 when using@latestbecause it constructed/releases/tags/latestinstead of using GitHub's/releases/latestAPI endpoint. This was a regression introduced in v2026.3.5. #8532 by @jdxFish shell shim PATH ordering on re-source -- When
config.fishis re-sourced (e.g., in VS Code integrated terminals),mise activate fish --shimsnow correctly moves shims to the front of PATH usingfish_add_path --global --move, instead of silently skipping them because they were already present. Other shells are unaffected. #8534 by @jdxTask output prefix disambiguation -- When running the same task multiple times with different arguments (e.g.,
mise run greet alice ::: greet bob), output prefixes now include the arguments to distinguish runs ([greet alice]vs[greet bob]). Arguments are only included when disambiguation is needed; single-instance tasks keep clean prefixes. Long prefixes are truncated to 40 characters. #8533, #8536 by @jdxNon-MRI Ruby on Windows -- Requesting non-MRI Ruby engines (jruby, truffleruby, etc.) on Windows now fails early with a clear error message explaining that only standard MRI Ruby is supported via RubyInstaller2, instead of producing a confusing 404 from an invalid download URL. #8539 by @jdx
Added
tigerbeetle(github:tigerbeetle/tigerbeetle) to the tool registry. #8514 by @risu729Breaking Changes
envtags withinmise.lockfor environment-specific version pinning, runningmise lockwill migrate to the new format automatically, creating separatemise.<env>.lockfiles. Make sure to commit the new lockfiles and update your.gitignoreif needed formise.<env>.local.lockfiles.touch_outputsremoved: If you were usingtouch_outputsin prepare provider configuration, that field is no longer recognized. Freshness is now determined entirely by blake3 hashing of source files.New Contributors
Full Changelog: jdx/mise@v2026.3.5...v2026.3.6
v2026.3.5: : Provenance tracking in lockfiles and task deduplication fixCompare Source
This release adds supply-chain security improvements by recording provenance verification results in lockfiles, exposes libc variant detection to vfox plugins, and fixes several bugs including duplicate task execution, offline mode hangs, and Windows binary identification.
Highlights
run = [{ task }].mise env,hook-env,activate, andexecfrom hanging when resolving"latest"versions behind private registries.Added
Provenance verification results stored in lockfiles --
mise locknow records which provenance mechanism (SLSA, GitHub attestations, cosign, or minisign) was used to verify each tool per platform. On subsequent installs, mise refuses to proceed if the recorded verification mechanism is disabled or unavailable, protecting against downgrade/stripping attacks. The lockfile format also changes from inline tables to dotted-key subtables for platform entries, improving readability. Existing lockfiles remain backwards-compatible and will be updated on the nextmise lock. #8495 by @jdxRUNTIME.envTypefor vfox plugins -- Vfox Lua plugins can now checkRUNTIME.envTypeto determine the libc variant at runtime ("gnu"for glibc,"musl"for musl Linux,nilon non-Linux). This lets plugins select the correct binary variant for the host system. #8493 by @maleptRegistry:
portless-- Addedportless(npm:portless) to the tool registry. #8508 by @risu729Fixed
Shared dependency tasks no longer run multiple times with task delegation -- When a task uses
run = [{ task }]to delegate, the sub-graph now inherits knowledge of tasks already completed in the parent graph, preventing shared dependencies from executing more than once. #8497 by @vadimpiven"latest"version no longer triggers network calls in prefer-offline mode --mise env,hook-env,activate, andexecwithprefer_offlineenabled would still make a remote call to resolve"latest"versions (e.g.,npm:pkg = "latest"). If the registry held the connection open waiting for credentials, mise would hang indefinitely. This is now skipped, matching the existing offline guard for fully-qualified versions. #8500 by @jdxWindows: mise binary correctly identified without
.exeextension -- On Windows,argv[0]can resolve tomise(without.exe),mise.bat, ormise.cmd, all of which were incorrectly treated as shims. This causedmise --helpandmise --versionto silently fail in some environments (e.g., conda-forge CI). A unifiedis_mise_binary()helper now handles all these variants. #8503 by @jdx, with credit to @salim-b for identifying the issue in #8496Full Changelog: jdx/mise@v2026.3.4...v2026.3.5
v2026.3.4: : Runtime musl detection, interactive tasks, and platform install fixesCompare Source
A feature-rich release that adds runtime musl/glibc detection for correct binary selection on Linux, a new
interactivetask field for exclusive terminal access, and several important fixes for platform-specific tool installation, the standalone installer, and Ruby precompiled binary discovery.Highlights
interactivetask field provides a targeted way to give a task exclusive terminal access without forcing all tasks to run sequentially.MISE_INSTALL_PATHpoints to an existing directory.Added
interactivefield for tasks -- Mark a task withinteractive = trueto give it exclusive terminal access (stdin/stdout/stderr) while other non-interactive tasks continue running in parallel. This is a more targeted alternative toraw = true, which forcesjobs=1globally --interactiveonly blocks concurrent tasks while the interactive task is actively running. #8491 by @jdxRuntime musl/glibc detection for correct libc variant selection -- mise now detects musl libc at runtime (by checking for
/lib/ld-musl-*) instead of using compile-time configuration. This means a musl-built mise running on a glibc system (or vice versa) will correctly select the right binary variant. Lockfiles now include separate entries forlinux-x64-muslandlinux-arm64-muslplatforms. Existing lockfiles without musl entries continue to work and will be updated on the nextmise lock. #8490 by @jdxHeader comment in generated lockfiles --
mise.lockfiles now include a@generatedheader comment, making it clear the file is auto-generated and should not be edited manually. #8481 by @ivyPreserve
.exeextensions on Windows -- The github, gitlab, forgejo, and http backends now automatically keep executable extensions (.exe,.bat,.cmd) when usingbinorrename_exeoptions on Windows, fixing tools like yt-dlp that were broken by extension stripping. #8424 by @ikiFixed
Registry platform options now applied during install -- Platform-specific options like
asset_patterndefined in the tool registry were silently ignored during installation because nested TOML structures were flattened to strings. This caused tools like flyway to select the wrong asset (e.g., alpine instead of linux-x64). #8492 by @jdxTool opts stored as native TOML to fix platform switching -- Switching an
http:tool from a single URL to platform-specific URLs ([tools."http:X".platforms]) could fail because cached options in.mise-installs.tomlwere mangled during round-tripping. Options are now stored as proper TOML fields with automatic migration of old manifests. #8448 by @jdxInstaller errors if
MISE_INSTALL_PATHis a directory -- SettingMISE_INSTALL_PATHto an existing directory (e.g.,~/tmpinstead of~/tmp/mise) caused the installer torm -rfthat directory, potentially deleting important files. The installer now detects this and exits with a clear error message suggesting a file path. #8468 by @jdxPrepare sources/outputs resolve relative to
dir-- When a prepare provider setsdir, relative source and output paths now correctly resolve againstproject_root/dirinstead of justproject_root. This fixes freshness tracking in monorepo setups where prepare providers target subdirectories. #8472 by @jdxRuby precompiled binary lookup for older versions -- Precompiled Ruby discovery used paginated release listing (first page only), so versions beyond the first 30 releases (like Ruby 3.2.2) silently fell back to compiling from source. The lookup now fetches the specific release by tag directly. #8488 by @jdx
JSON schema supports structured objects in task depends -- The JSON schema for
depends,depends_post, andwait_fornow correctly accepts the structured{ task, args?, env? }object syntax that the runtime already supported, fixing IDE validation errors. #8463 by @risu729Broken pipe no longer panics in task output -- Task output macros used
println!/eprintln!which panic on broken pipes (e.g., when piping mise output tohead). Replaced withcalm_ioequivalents that gracefully handle closed stdout/stderr. #8485 by @vmalezeScoped npm package names no longer panic -- Using
@scope/pkg(e.g.,@anthropic-ai/claude-code) without thenpm:backend prefix caused an internal panic. The parser now correctly treats the leading@as part of the package name and provides a proper error message. #8477 by @jdxNew Contributors
Full Changelog: jdx/mise@v2026.3.3...v2026.3.4
v2026.3.3: : Standalone installer zstd fixCompare Source
A single-fix patch release that corrects the standalone installer's zstd archive selection logic on systems where the
zstdbinary is not installed.Fixed
zstdis not installed -- Thetar_supports_zstd()function returnedtruefor GNU tar >= 1.31 regardless of whether thezstdbinary was actually present on the system. Since GNU tar shells out tozstdrather than linking against it, this caused extraction to fail. The fix checks for thezstdbinary upfront before evaluating the tar version. A separate fallback branch inget_ext()that could also select a.tar.zstarchive without verifying zstd availability has been removed. #8460 by @octoNew Contributors
Full Changelog: jdx/mise@v2026.3.2...v2026.3.3
v2026.3.2: : Local-scoped upgrades, config-based quiet/silent, and redaction fixesCompare Source
A small release that adds a
--localflag for scopingoutdatedandupgradeto project-local tools, fixes several bugs around env var redaction, Tera template rendering in prepare, and task output configuration, and corrects a regression in the standalone installer.Added
--localflag foroutdatedandupgradecommands -- You can now runmise upgrade --localormise outdated --localto restrict operations to tools defined in project-local config files (e.g.,mise.toml), skipping anything from the global config (~/.config/mise/config.toml). This is useful when you have separate workflows for managing global vs. project-local tool versions. #8451 by @malepttinygo added to the registry -- TinyGo is now available via
mise use tinygo, using the aqua backend. #8446 by @artemklevtsovFixed
task.outputconfig setting now works for quiet/silent modes -- Settingtask.output = "quiet"ortask.output = "silent"inmise.tomlwas not suppressing mise's own output during task runs. Only theMISE_TASK_OUTPUTenvironment variable worked. Both approaches now behave identically. #8445 by @my1e5Redactions applied correctly when
tools = trueandredact = trueare combined -- Environment variables with bothtools = trueandredact = truewere not being redacted in task output because the tools-only code path collected redactions but never registered them with the global redactor. Secret values now correctly appear as[redacted]. #8449 by @jdxTera templates rendered in
[prepare.*.env]values -- Tera template expressions like"{{env.MY_VAR}}"in prepare provider env blocks were being passed as literal strings instead of being rendered. They are now evaluated with the full toolset environment available in the template context. #8450 by @jdxStandalone installer tar zstd version check regex fixed (again) -- A follow-up fix to the regex repair in v2026.3.1: a missing escape on the opening parenthesis caused
grep: Unmatched ) or \)errors. The regex is now fully correct. #8453 by @chadlwilsonFull Changelog: jdx/mise@v2026.3.1...v2026.3.2
v2026.3.1: : Bug fixes for tasks, Swift, Julia, and installerCompare Source
A small bug-fix release that corrects task scheduling behavior for
depends_post, fixes Julia version resolution, resolves Swift installation failures on some macOS environments, and repairs the standalone installer's tar version detection.Fixed
depends_posttasks no longer run when a pre-dependency fails -- Previously, if a task independsfailed before the main task started,depends_postcleanup tasks would still execute. This happened because the scheduler checked whether the parent task was scheduled rather than whether it actually executed. Nowdepends_posttasks are correctly skipped when the main task never ran due to a pre-dependency failure. They still run as expected when the main task itself fails. #8426 by @jdxJulia version listing no longer crashes with
MISE_USE_VERSIONS_HOST=0-- The Julia registry entry'sversion_exprused afilter()predicate with a bare#variable, which the expr-lang evaluator treated as an undefined variable. Wrapping the predicate in{...}closure braces fixes the syntax somise ls-remote juliaworks correctly when fetching versions directly from the upstream JSON endpoint. #8420 by @jdxSwift install fallback to system
pkgutilon macOS -- On some macOS environments,pkgutilis not found on the PATH during Swift installation. The Swift backend now falls back to/usr/sbin/pkgutil(the standard system location) whenwhich pkgutilfails. #8435 by @mackwicStandalone installer tar zstd version check fixed -- The regex used to detect whether the system
tarsupports zstd was broken -- unescaped parentheses and pipes causedgrepto match nothing. The regex is now properly escaped so tar version 1.31+ is correctly detected, enabling zstd-compressed archive downloads. #8430 by @autarchNew Contributors
Full Changelog: jdx/mise@v2026.3.0...v2026.3.1
v2026.3.0: : Smarter prepare, task-backed hooks, and per-task varsCompare Source
This release brings a major upgrade to
mise preparewith content-hash freshness, dependency ordering, and better diagnostics. Hooks and watch files can now delegate to full mise tasks, and taskvarsgain monorepo inheritance and per-task overrides. Several lockfile and idiomatic version file parsing bugs are also fixed.Highlights
mise prepareoverhaul -- Freshness detection switches from mtime to blake3 content hashing (reliable across CI and clock skew), providers can declare dependencies on each other, a new--explainflag shows detailed diagnostics, and per-provider timeouts are now supported.watch_filescan now reference mise tasks instead of inline scripts, gaining access to the full task system (deps, env, templating).varsthat override config-level vars, and monorepo subdirectory vars are now properly inherited when running tasks from the root.Added
Task references in hooks and watch_files -- Hooks can now use
{ task = "name" }syntax to run a mise task instead of an inline script. Mixed arrays of scripts and task references are supported. Task refs respectMISE_NO_HOOKS=1and the full task system (deps, env, etc.). #8400 by @jdxPer-task vars and monorepo vars inheritance -- Tasks can now define task-local
varsthat override config-level vars for that task. Monorepo subdirectory vars are also properly inherited when running tasks from the project root, matching howenvalready works. #8248 by @halmsBuilt-in
git-submoduleprepare provider -- A new built-in provider detects.gitmodulesand runsgit submodule update --init --recursivewhen submodule directories are stale. No configuration needed -- it activates automatically when.gitmodulesexists. #8407 by @jdxBlake3 content-hash freshness for prepare --
mise preparenow uses blake3 content hashing instead of mtime comparison to determine whether providers need to run. Hashes are persisted to.mise/prepare-state.toml. This is more reliable across CI, VCS operations, and clock skew scenarios. #8404 by @jdxHuman-readable stale reasons in prepare output -- Dry-run and status bar messages now explain why a provider would run, e.g.
[dry-run] Would prepare: npm (node_modules does not exist)orprepare: codegen (schema.graphql changed) -- run 'mise prep'. #8408 by @jdxmise prepare --explain <provider>diagnostics -- A new--explainflag shows detailed information about a specific provider: sources, outputs, auto status, command, and a fresh/stale verdict with reasons. Exits 0 if fresh, 1 if stale, useful for scripting. #8409 by @jdxPer-provider timeout support for prepare -- Providers can now set a
timeoutin seconds. If the command exceeds the timeout it is killed. #8405 by @jdxDependency ordering for prepare providers -- Providers can declare
depends = ["other-provider"]to enforce execution ordering. Independent providers still run in parallel. Cycle detection, failure propagation, and unknown-dep warnings are all handled. #8401 by @jdxFixed
Idiomatic version files now ignore comments -- Files like
.python-version,.node-version,.ruby-version, and.java-versionnow correctly strip#comments (both full-line and inline) and blank lines, preventing mise from treating comment text as version specifiers. #7682 by @iloveitalyGeneric parser used for idiomatic files -- Built-in idiomatic file parsers (e.g. for
package.jsonand raw text files) are now used consistently, preventing unexpected behavior when plugins like vfox don't support certain file formats. #8171 by @risu729Aqua bin_paths disk cache restored with proper invalidation -- The aqua
bin_paths.msgpack.zcache removed in v2026.2.24 has been restored withfresh_fileinvalidation keyed on the install directory. This recovers the 7-11% performance regression onmise envandmise hook-envwhile keeping cache correctness. #8398 by @jdxLockfile no longer splits entries for precompiled settings --
mise lockwithprecompiled_flavoror similar settings configured no longer produces duplicate tool entries by splitting the host platform into a separate entry. #8396 by @jdxPython lockfile respects precompiled settings --
mise locknow correctly usesprecompiled_arch,precompiled_os, andprecompiled_flavorsettings when generating Python lock file entries, andprecompiled_flavoris properly honored during installs. #8399 by @jdx"v" prefix normalized in lockfile version matching --
--lockedmode no longer fails whenmise.tomlspecifiesv1.2.3but the lockfile stores1.2.3(or vice versa). #8413 by @jdxVfox no longer eagerly loads metadata -- Removed vfox's
idiomatic_filenames()override that triggered plugin metadata loading for every config file check, and reordered detection logic so known patterns are checked first. Eliminates spurious[vfox] Getting metadata for yarndebug messages. #8397 by @jdxFixed infinite recursion with
uv_venv_autoand uv shims -- Whenuv_venv_auto = "create|source"is set and a mise shim foruvexists on PATH, venv creation no longer enters infinite subprocess recursion. The fix excludes the mise shims directory from theuvbinary search. #8402 by @halmsImproved git submodule parser for prepare -- The
.gitmodulesparser is now INI-section aware, only extractingpathvalues from[submodule "..."]sections and ignoring comments. Freshness check errors now default to fresh rather than stale, preventing spurious warnings. #8412 by @jdxBreaking Changes
# mise ...file task header syntax has been removed as scheduled. Only#MISE/# [MISE]///MISE/::MISEheaders are now recognized. If you have task files still using the old# miseheaders, update them to use the new syntax. #8403 by @jdxNew Contributors
Full Changelog: jdx/mise@v2026.2.24...v2026.3.0
v2026.2.24: : Hooks get Tera templates, aqua cache cleanup, and better error messagesCompare Source
A bug-fix release that enables Tera template rendering in hooks, eliminates a class of stale PATH bugs with aqua tools, improves error messages for unsupported registry tools, and removes the long-deprecated
python.venv_auto_createsetting.Fixed
Hooks now support Tera template rendering -- Hook scripts can now use Tera template variables like
{{tools.ripgrep.path}}, just like tasks. Additionally, the install progress bar is now cleared before postinstall hooks run, so hook output is no longer masked by the progress UI. Preinstall hooks correctly skiptools=trueenv directives since referenced tools may not yet be installed. #8385 by @jdxAqua tool PATH entries no longer go stale -- The aqua backend's
bin_pathsdisk cache (bin_paths.msgpack.z) has been removed entirely. This cache provided negligible performance benefit -- reading and decompressing a msgpack file is comparable to parsing the small YAML registry entry -- but was the root cause of stale PATH entries after tool installs (e.g. upgradinguvcausing its PATH entry to vanish). The previous fix in v2026.2.23 was raceable by concurrentmise hook-envcalls; removing the cache eliminates this class of bugs completely. #8383 by @jdxBetter error when a registry tool has no supported backends -- When a tool exists in the registry but all its backends are filtered out for the current platform or configuration (e.g.
imagemagickon a platform where onlycondaandasdfbackends are registered but disabled), the error now clearly explains the situation and lists the registered backends instead of suggesting the user meant the exact tool name they already typed. #8388 by @jdxRemoved
python.venv_auto_createsetting removed -- Thepython.venv_auto_createandpython_venv_auto_createsettings have been fully removed. These were deprecated in favor of the_.python.venvconfiguration. If you were relying on the legacyvirtualenvtool option to auto-create venvs, mise will now warn and print manual creation instructions instead. Migrate to the newer venv configuration: #8384 by @jdxBreaking Changes
python.venv_auto_createandpython_venv_auto_createsettings no longer exist. If you still have these in your configuration, they will be silently ignored. Use_.python.venv = { path = ".venv", create = true }in your tool configuration instead. #8384Full Changelog: jdx/mise@v2026.2.23...v2026.2.24
v2026.2.23: : Stricter lockfile enforcement and vfox backend optionsCompare Source
This release tightens lockfile behavior in
--lockedmode, fixes a stale PATH cache issue with aqua-based tools, resolves intermittent panics with remote git tasks, and adds the ability to pass custom options to vfox backend plugins.Added
Custom options for vfox backend plugins -- Options defined in
mise.tomltool entries are now passed through to vfox backend plugins in bothBackendInstallandBackendExecEnvcontexts, accessible in Lua viactx.options. This enables custom plugin use cases like controlling build parameters. #8369 by @Attempt3035Registry: porter -- Added Porter, a CNAB bundle authoring and management tool (
github:getporter/porter). #8380 by @lbergnehrRegistry: entire -- Added entire CLI (
aqua:entireio/cli). #8378 by @TyceHerrmanRegistry: topgrade -- Added topgrade (
aqua:topgrade-rs/topgrade), an all-in-one system upgrade tool. #8377 by @TyceHerrmanFixed
--lockedmode now strictly enforces the lockfile -- Previously,mise lockcould still run while--lockedwas active,mise use tool@latestcould bypass the lockfile, and tools missing from the lockfile would silently fall through to remote resolution. Nowmise lockrefuses to run in locked mode with a clear error and hint,mise use tool@latestrespects the lockfile when locked, and missing tools fail fast with an actionable message instead of resolving remotely. #8362 by @jdxAqua tool PATH entries no longer go missing after install -- The
list_bin_paths()cache could be populated with stale (empty) data before extraction finished, or by a concurrentmise hook-envcall during installation. The in-memory and on-diskbin_pathscaches are now cleared after an aqua tool install completes so paths are recomputed from the freshly installed files. Fixes an issue where upgrading tools likeuvcaused their PATH entry to vanish. #8374 by @jdxRemote git task cache no longer panics or corrupts on concurrent access -- Replaced
println!/eprintln!with non-panickingwriteln!to handle EPIPE gracefully, and added file locking with clone-to-temp-then-rename to prevent concurrent cache corruption when multiple mise processes fetch the same remote git task simultaneously. #8375 by @vmalezeNew Contributors
Full Changelog: jdx/mise@v2026.2.22...v2026.2.23
v2026.2.22: : Outdated plugins, rename_exe fixes, and smoother installsCompare Source
A small release adding a new way to check for outdated plugins, along with three bug fixes for archive installs, tool environment resolution, and cross-platform Ruby lockfiles.
Added
mise plugins ls --outdatedflag -- A new-o/--outdatedflag checks remote git refs in parallel and displays only plugins where the local SHA differs from the remote. Shows a table with plugin name, URL, ref, local SHA, and remote SHA. Prints "All plugins are up to date" when everything is current. #8360 by @jdxFixed
rename_exeworks with archives containing abin/subdirectory -- When an archive extracts to a layout likeprefix/bin/binary, therename_exeoption was silently skipped because it searched the extraction root non-recursively instead of thebin/subdirectory where the binary actually lives. Both the GitHub-style backend and the HTTP backend now auto-detect thebin/subdirectory as the search directory, matching the same logic used bydiscover_bin_paths()for PATH construction. #8358 by @jdxInstalling cargo/npm/pipx tools no longer crashes with
tools = trueenv directives -- When[env]contained entries likeNODE_VERSION = { value = "{{ tools.node.version }}", tools = true }, installing npm/cargo/pipx tools would fail with "Variable not found in context" because the referenced tools might not be installed yet. The cargo, npm, and pipx backends now skiptools = trueenv directive resolution during installation while still including tool paths in PATH. #8356 by @jdxRuby lockfile resolves correct Windows checksums -- Running
mise lockon macOS/Linux now correctly resolves RubyInstaller2 binary URLs and checksums for Windows platform entries, instead of incorrectly using the MRI source tarball checksum. The lockfile generator now fetches the correct.7zasset from theoneclick/rubyinstaller2GitHub releases for Windows targets. #8357 by @jdxChanged
github:chenrui333/terradozer(replacing the archived asdf plugin and unavailable aqua backend), and is restricted to Linux and macOS. #8365 by @chenrui333New Contributors
Full Changelog: jdx/mise@v2026.2.21...v2026.2.22
v2026.2.21: : Bug fix roundup for monorepo tasks, conda noarch, and exec PATH handlingCompare Source
A bug-fix release addressing several regressions and long-standing issues: monorepo task variables and glob dependencies now resolve correctly, the conda backend can install Python noarch packages,
mise xrespects virtualenv PATH ordering again, and nested task execution no longer hangs.Fixed
mise xrespects virtualenv PATH order again -- A pre-resolution step added in v2026.2.17 (#8276) resolved bare command names directly to mise-managed tool paths, bypassing PATH entirely. This broke_.python.venvand similar configs where a virtualenv binary should take precedence over the mise-managed install. The pre-resolution is removed; shim stripping inexec_program(also from #8276) is sufficient to prevent recursion. #8342 by @jdxConda noarch Python packages install correctly -- Installing noarch Python packages via the conda backend (e.g.
mise use conda:ruff) failed because the linker didn't know the Python version needed to computesite-packagespaths. The solver's resolved Python version is now extracted and passed through tolink_package, fixing the error. #8349 by @wolfvNested mise tasks no longer hang -- The process group isolation (
setpgid/killpg) introduced in v2026.2.18 and refined in v2026.2.19 has been fully reverted. When tools like Playwright use process-group-based kills (kill(-pid, SIGKILL)) to tear down a server subprocess tree, grandchild processes in a separate group (created by mise'ssetpgid) survived and held pipes open, causing indefinite hangs. mise now sends signals directly to child PIDs instead. #8347 by @jdxMonorepo tasks resolve
[vars]from subdirectory configs -- Running a monorepo task likemise run //infra/stacks/gcp:greetfailed to pick up[vars]defined in subdirectory.mise.tomlfiles, causing template rendering errors. Variables are now resolved from the task's full config hierarchy (includingmise.<env>.tomloverlays) and threaded through script rendering. #8343 by @jdxMonorepo glob dependencies trigger subdirectory prepare steps -- When a root task depended on a monorepo glob pattern like
//...:check, the prepare phase only collected configs from top-level tasks, missing subdirectory tasks entirely. Dependencies are now resolved before prepare runs, so transitive subdirectory tasks and their prepare providers are properly discovered. #8353 by @jdxNew Contributors
Full Changelog: jdx/mise@v2026.2.20...v2026.2.21
v2026.2.20: : Conda rewrite, .NET SDK core plugin, and per-task timeoutsCompare Source
A feature-packed release that replaces the conda backend with production-grade internals, adds a native .NET SDK plugin, and finally enforces per-task timeouts. Several lockfile and environment-handling fixes round things out.
Highlights
Conda backend rewritten with rattler -- The experimental conda backend has been completely rewritten to use the rattler Rust crates (the same engine behind pixi), replacing ~1,600 lines of custom code that relied on the unsupported anaconda.org API. This brings a proper SAT-based dependency solver, correct binary prefix replacement, and repodata caching via CDN. #8325 by @jdx
Native .NET SDK management -- A new core plugin for .NET SDK installs all versions side-by-side under a shared
DOTNET_ROOT, matching .NET's native multi-version model. It uses Microsoft's officialdotnet-installscript and supportsglobal.jsonfor per-project SDK pinning. #8326 by @jdxPer-task timeouts are now enforced -- The
timeoutfield on tasks (added in v2025.1.6 but never wired up) now actually kills tasks that exceed their limit. Timeouts send SIGTERM with a 5-second grace period before SIGKILL, and both per-task and globaltask_timeoutsettings are respected. #8250 by @tvararuAdded
mise use dotnet@8now installs via a native core plugin with side-by-side version support andglobal.jsondetection. ConfigureDOTNET_ROOTvia the newdotnet.dotnet_rootsetting. #8326 by @jdxtimeoutfield are now killed if they exceed the configured duration. Works with both per-task config and the globaltask_timeout/--timeoutflag. #8250 by @tvararu.vsixfiles as ZIP archives and extracts them correctly, enabling tools distributed as VS Code extensions to be installed viahttp:URLs. #8306 by @sosumappuoxfmt-- Added the oxfmt formatter to the tool registry. #8316 by @taoufik07Changed
Configuration
📅 Schedule: Branch creation - At any time (no schedule defined), Automerge - At any time (no schedule defined).
🚦 Automerge: Disabled by config. Please merge this manually once you are satisfied.
♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.
🔕 Ignore: Close this PR and you won't be reminded about this update again.
This PR was generated by Mend Renovate. View the repository job log.