Repository navigation
Release 0.5.0
π Features
- feat(clickhouse): log the server to stdout so pod logs carry it (4b71d59)
- feat(compute-kubernetes): tell when an instance's data volume is gone (534bd43)
- feat(storage):
gfs storage reclaimasks before deleting when run at a terminal (187ecb7) - feat(storage):
gfs storage reclaimdeletes the OpenEBS ZFS volumes the driver was asked to delete and never did (2ca5aa2) - fix(k8s-checkout): confirm the VolumeSnapshot before destroying the live volume, and before moving HEAD (dbc2f59)
- feat(k8s): a start can apply resources to an instance built without them (fb508ea)
- feat(compute): find the instances provisioned for a repository (4031463)
- feat(compute): an instance epoch that changes whenever the instance starts or stops (032f0ba)
- feat(compute-kubernetes): tell a database scaled to zero from one whose StatefulSet is gone (4681b8e)
- feat(gfs-postgres): offer TLS on every start (e169c67)
- fix(repo): add write_durable, an atomic write that also survives a crash (3f788bc)
- fix(config): write config.toml and credentials.toml atomically (eb4c28c)
- feat(fsck): verify an object hashes to its own address (538472d)
- fix(fsck): exit 1 only when something can actually be collected, and stop a refusal from borrowing that code (04c3ac6)
- feat(fsck): prove a detached working copy is safe to remove, or say why not (cb66d53)
- fix(fsck): a branch deleted before its first commit is not corruption (f31ba7a)
- feat(fsck): report what deleting a snapshot would actually free, on Kubernetes (891d1f6)
- feat(fsck): report unreachable and dangling objects without removing anything (b54621a)
- feat(status): surface whether snapshots share extents or copy bytes (de44c9a)
- feat(k8s): report what the node has left, so admission can refuse a full node (2401e65)
- feat(gfs): record the resource spec with the repository so rebuilds re-apply it (da7a6e7)
- feat(k8s): give the container real resource requirements, and let definition labels reach the pod (0887262)
- feat(domain): carry an optional resource declaration on ComputeDefinition (7eaa1b4)
- feat(domain): add a ComputeResources declaration type (45444cc)
- feat(branch): make
branch -drecoverable instead of silently destructive (9733af6) - feat(clone): bootstrap the lazy clone in-pod, over loopback (93ed8b2)
- feat(compute-kubernetes): detect pod readiness with a watch, not a 2s poll (f5a62fb)
- feat(compute-kubernetes): stream pod readiness via a watch instead of only polling (c8b8faf)
- feat(compute-kubernetes): report Running only when serving; complete instance destroy (9703bfb)
- feat(sqlite): export and import in-process, and document the provider (d1b15c2)
- feat(cli): stop requiring a container runtime for embedded providers (9d8a07a)
- feat(domain): route embedded providers past the compute instance (1ee5aed)
- feat(sqlite): add file-based SQLite provider with in-process engine (80fe4cb)
- feat(users): reassign a dropped role's objects to the deploy owner (0427661)
- feat(compute): honor GFS_CONTAINER_UID_GID to override the container run-as user (9188890)
- feat(clone): snapshot mode, remote-sync verbs, and torn-clone detection (d49f696)
- feat(clone): adopt new partitions, inherited children and matview refreshes (efe26e9)
- feat(clone): notice when the source adds a constraint (49344b9)
- feat(clone): handle source schema changes instead of failing on raw SQL (ffee385)
- feat(clone): self-heal a stale table instead of federating it forever (4720f14)
- feat(clone): stop serving stale rows when the source has changed (00d8a70)
- feat(clone): detect when the source changes after a clone is created (0d5333a)
- feat(db-user): record + surface the applied preset in list (0c44eae)
- feat(vault): add RepoCredentialVault (.gfs/secrets node-local store) (dc30eba)
- feat(db-user): gfs user grant/revoke/list-privs CLI + MCP tool (story 007-004) (8b889e7)
- feat(db-user): object-level GRANT/REVOKE + list-privileges (Postgres) (2461cbd)
- feat(user): add deploy-env bootstrap primitive (RFC 009 story 1) (5d55dbc)
- feat(user): add gfs user MCP tool (create/list/drop/set-password) (742be9b)
- feat(user): add gfs user CLI (create/list/drop/set-password) + fix cmd_log sort lint (e9c46d7)
- feat(user): gfs user domain core β RolePreset/RoleSpec, hardened provider role SQL, ManageUsersUseCase (5b267ea)
- feat(compute): add Compute::read_deploy_credentials (backend-agnostic) (e5bdc11)
- feat(cli): add 'gfs destroy' to tear down an instance and remove .gfs (b3b94b0)
- feat(storage-k8s): use cluster-default VolumeSnapshotClass when unset (harden 2/N) (400f107)
- feat(gfs-postgres): parameterize base image + add offline build recipes (882a955)
- feat(storage-kubernetes): report real ZFS volume usage in status() (9532181)
- feat(postgresql): update default image to gfs-postgres:17 and enhance image tagging logic (008dcaf)
- feat(init): persist display_name on repo init (ffd7fff)
- feat(clone): detach remote SQL for snapshot clones (fcf8b4b)
- feat(clone): k8s lazy clone and rebase fixes (a83aeb4)
- feat(cli): pass project env var on init (fd847e8)
- feat(cli): remote runtime mode (6730b23)
- feat: expose list_branches for repo graph API (138f789)
- feat(k8s): CP branching via stable PVC restore (4e2f389)
- feat(cli): remote runtime support on k3s (58aac62)
- feat(cli): k8s branch/query/checkout e2e proof (0abe5ea)
- feat: add k3s/kubernetes runtime for GFS (5f8e219)
π Bug Fixes
- fix(mysql): build JSON booleans in schema extraction (4a4acf8)
- fix(commit): capture a live database before reading its schema (ebf6c1f)
- fix(compute-kubernetes): provision an unexposed instance on ClusterIP, with no NodePort or host port at any time (c9e723e)
- fix(extension): hydrate as a restricted operation on a pinned search path, and keep coverage writes off PUBLIC (7fee14e)
- fix(domain): quiesce a database through its internal address, never its published port (edc0675)
- fix(extension): hydrate a lazy table as its owner, without letting the reader set the replication role (2eca6f9)
- fix(branch): drop the unreachable guard #180 left in branch -d, and say why (92fc5f4)
- fix(test): assert on what gfs checkout printed, not on what libtest printed (ddad01a)
- fix(checkout): a read after a commit is not uncommitted work (2fd0e19)
- fix(cli): route branch delete's detached-workspace notice through println_safe! (9de7298)
- fix(k8s-destroy): a cluster without the VolumeSnapshot CRD has no snapshots, not an error (d0ed1ff)
- fix(k8s-destroy): finish the ZFS volume deletes OpenEBS defers, so a destroyed database's data leaves the node (fc45726)
- fix(k8s-checkout): confirm the VolumeSnapshot before destroying the live volume, and before moving HEAD (dbc2f59)
- fix(cli): --help and --version never panic on a closed reader (f20ba9f)
- fix(cli): route status's copy-on-write notice through println_safe! (391c85a)
- fix(cli): a closed reader never panics a command (a72a9d9)
- fix(cli): a non-UTF-8 argument is a usage error, not a crash (ef55f24)
- fix(fsck): write the --plan file atomically and durably (6e5e3fb)
- fix(fsck): a leftover temp from an interrupted write is not corruption (d5495b3)
- fix(compute-kubernetes): an exec into an instance whose StatefulSet is gone fails at once (cd1aec1)
- fix(postgres): send role passwords as SCRAM verifiers, never plaintext (663bcd0)
- fix(postgres): pass inline psql SQL as one single-quoted word (97804f1)
- fix(commit): discard the snapshot of a commit whose schema capture failed (472bf8c)
- fix(commit): a stopped database is committed with the schema captured when it stopped (ce884f5)
- fix(commit): skip the quiesce only for a database known to be stopped (78f9fd1)
- fix(clone): the remote version probe reports why it failed on Kubernetes (8cc24c6)
- fix(domain): init records the provisioned instance before starting it (60c6dc1)
- fix(sqlite): stop checkpointing the WAL before a snapshot, which corrupts it (8510dcc)
- fix(tests): stop a gfs-postgres image built for the wrong PostgreSQL from passing as the right one (59e3244)
- fix(repo): skip a taken temp name instead of deleting someone else's file (ae4eb90)
- fix(repo): write refs, HEAD, WORKSPACE and objects durably; skip leftover temps (9e5fb51)
- fix(repo): add write_durable, an atomic write that also survives a crash (3f788bc)
- fix(config): write config.toml and credentials.toml atomically (eb4c28c)
- fix(checkout): flush once before comparing, not once per comparison (6092122)
- fix(fsck): date a workspace by the newest write in it, not its container (c3bf133)
- fix(branch): the reserved name is matched case-insensitively, by segment (7e1b2a7)
- fix(branch): 'detached' is a reserved name, and checkout -b now validates like branch (1b26050)
- fix(checkout): stop reporting the database's own bookkeeping as uncommitted work (5683650)
- fix(fsck): close the guards the code already decided it wanted, and reconcile the docs (a741c13)
- fix(fsck): an empty HEAD is a hole, and safe_to_remove stops claiming content (acccb42)
- fix(docker,commit): stop redirecting PGDATA on colima, and let a stopped database be committed (575342a)
- fix(docker): refuse a bind mount the container runtime's VM does not share (116b622)
- fix(docker): do not pin a container to the host uid when the bind root is root-owned (a191e40)
- fix(cli): a usage error exits 3, not the code that means "ran and found something" (6603ee0)
- fix(fsck): make three messages describe what is actually wrong (275b1a5)
- fix(log): stop
gfs log --jsonpanicking when the reader has gone (51d1ffe) - fix(fsck): exit 1 only when something can actually be collected, and stop a refusal from borrowing that code (04c3ac6)
- fix(fsck): compare workspace paths as directories, not as spellings (cbfaebd)
- fix(fsck): derive the snapshot owner prefix the way the label was written (708f5ab)
- fix(fsck): a refusal reports the repository's own verdict, on the asked-for channel (16c9d93)
- fix(fsck): the persisted plan records the epoch the mark ran against (4748514)
- fix(fsck): read the branch store twice, so a concurrent restore cannot hide a branch (a53e748)
- fix(cli): a closed reader must not replace the command's verdict (3d1e0a6)
- fix(fsck): absence and unreadability are holes, not an empty root set (4cd0d0b)
- fix(fsck): put a floor under --grace, and zero behind a flag that names itself (1d3d42c)
- fix(fsck): a branch deleted before its first commit is not corruption (f31ba7a)
- fix(domain): resolve the recorded workspace against the repository, not the cwd (c986852)
- fix(fsck): --json emits an error object when the check cannot run (98eb41f)
- fix(fsck): a shard we could not stat is not an unexpected entry (09b1b11)
- fix(fsck): an object we cannot open is a hole, not corruption (c599379)
- fix(fsck): an absent HEAD is a hole, not a repository with nothing detached (6f55d0b)
- fix(fsck): an unreadable refs/ is not a repository with no branches (6d439d1)
- fix(fsck): the CLI kept its own copy of the collapse the domain just lost (398e496)
- fix(fsck): give "I could not look" one channel instead of five silent collapses (6023d98)
- fix(fsck): a corrupt commit hides its ancestors rather than condemning them (062952f)
- fix(fsck): stop proposing other repositories' live snapshots for collection (682b5c7)
- fix(fsck): a broken ref no longer offers the whole repository up for collection (9debbe1)
- fix(fsck): treat soft-deleted branches as roots, or a collector eats them (4b7e860)
- fix(fsck): stop telling users a workspace is safe to delete β on main it is not (6ec395c)
- fix(fsck): report referenced bytes honestly, and verify snapshots on Kubernetes (1ab98c5)
- fix(fsck): stop the grace period hiding corruption, and repair the workspace walk (fc4a0a7)
- fix(fsck): stop the mark phase marking live data, and report stale working copies (b72004b)
- fix(docker): give every container the host alias, and skip the suites that need it when it is absent (8bb40a9)
- fix(docker): recognise colima's bind mounts, which need the same PGDATA redirect as podman (ee3f604)
- fix(commit,init): refuse a snapshot that captured nothing, and catch the cause at init (406d140)
- fix(k8s-checkout): re-apply the repository's discovery labels on a branch switch (8e36e18)
- fix(storage): probe clonefile on macOS too, where the same defect is unreported (af60cde)
- fix(storage): report when a snapshot will be a full copy, not a clone (b298479)
- fix(commit): say that an unfrozen snapshot may be silently lossy (fdf085b)
- fix(k8s): a recorded database_port pins one Service port, not every port (ba14b26)
- fix(k8s): the instance NodePort may pin one Service port, not every port (b77517a)
- fix(docker): detect the mount failure Docker Desktop actually reports (88f1e5b)
- fix(k8s-checkout): scope DatabaseProvider to the test that uses it (ef71be5)
- fix(mcp): require a bearer token on every HTTP transport (cbddced)
- fix(branch): validate branch names, which reach the filesystem unchecked (7017c48)
- fix(k8s-checkout): re-apply the repository's tuning parameters on a branch switch (4b409f8)
- fix(repo): resolve cp from a standard directory in copy_dir_all (59bec21)
- fix(tests): gate postgres readiness on TCP, the transport under test (b7d495b)
- fix(sqlite): assert the nested path with the platform separator (ac53767)
- fix(checkout): keep every component of a nested branch name (6f2362d)
- fix(checkout): catch the size-identical write, and stop a plain read reading as dirty (0418c3e)
- fix(compute): refuse
compute configon an embedded provider too (79a9684) - fix: refuse a missing snapshot before deleting the workspace, not after (7f9761b)
- fix(schema): name a schema object by everything it stores, and stop the sentinel commit id from panicking (1553986)
- fix: take the lock on the Kubernetes checkout, and stop naming the wrong command (45d51e1)
- fix(scripts): make the race harness able to fail (f4ebbff)
- fix: a concurrent commit and checkout could lose a commit (2d8f854)
- fix(refs): scope HEADS_DIR and REFS_DIR to the tests that still use them (9597d65)
- fix(refs): keep the symlink-escape test off Windows (48b7e78)
- fix(refs): refuse an existing ref at the write, not with a predicate (491d7d4)
- fix(mcp): judge a line blank only once its terminator has arrived (9902d2c)
- fix(storage): keep the two platform-bound tests off the platforms they cannot pass (bce4baa)
- fix(branch): give every deletion its own slot, not every name (27e1f37)
- fix(branch): enforce retention in the domain, and repair five smaller audit findings (c931c03)
- fix(branch): validate branch names, enforce retention on read, and refuse conflicting flags (5b1a9a5)
- fix(k8s): name a user and database in the readiness probe (bf30d80)
- fix(clone): keep the source password out of the snapshot, and pin the launcher for real (05ba192)
- fix(ci): record tokio-stream in the lock, and make CI fail on a stale one (10c43fa)
- fix(gfs): grant the planner hook the catalog writes it makes on a read (b553386)
- fix(users): never hand drop_role a superuser (e546dc9)
- fix(cli): announce an error once, not twice (eca9410)
- fix(rebase): keep main's section header wording (89243a1)
- fix(status): report the HEAD commit from one place, so the CLI and MCP cannot disagree (cd1f60c)
- fix: two silent losses a blind stress audit found, both mine (8f82093)
- fix(storage): resolve
cpin every backend, not just the one I noticed (9499666) - fix: three leftovers β a PATH-dependent commit, an unenforced invariant, and expression-index uniqueness (8b71084)
- fix(checkout): a branch could materialise content it never contained, and a detached commit could be lost (a254b8c)
- fix(sqlite): a symlinked database committed as success with zero data, and seven smaller findings (b2cf842)
- fix: a concurrent commit and checkout could lose a commit, and five smaller findings (6287457)
- fix(mcp): SQLite worked in the skill file and nowhere else (a919592)
- fix(sqlite): export corrupted data on restore β triggers fired, virtual tables emptied (e721e90)
- fix: quiesce embedded databases for schema capture and across checkout (6a05d0c)
- fix(sqlite): the resolver failed open β three silent paths to a wrong commit (9cf4a69)
- fix(sqlite): find the database instead of assuming it is called db.sqlite (8b1f23c)
- fix(sqlite): five ways the recorded schema misdescribed the database (f39dccc)
- fix(sqlite): stop reporting every snapshot failure as lock contention (cca8f48)
- fix(commit): quiesce embedded databases across the storage snapshot (83dd06a)
- fix(sqlite): report nullability and uniqueness as SQLite behaves (b226f98)
- fix(mcp): survive a blank line, answer argument errors one way, and name the actual condition when a directory is not a repository (0dc175a)
- fix(storage): resolve
cpto a standard location instead of trusting PATH (a5941e5) - fix(checkout): guard the workspace being rebuilt, not the one being left (58ebbc6)
- fix(checkout): restore the workspace from the snapshot, and refuse to overwrite uncommitted work (1f3404e)
- fix(commit): refuse to commit at a detached HEAD instead of orphaning (23a91db)
- fix(console): pass the snapshot arguments cmd_clone now takes (cb055b6)
- fix(build): reserve an 8 MiB main-thread stack for the Windows binary (24a4a39)
- fix(clone): report tables created on the source after clone time (91e60a5)
- fix(clone): re-sync enum labels, and stop hiding changes nothing accounts for (fdfe4cd)
- fix(clone): advance local sequences to the source's position (29756b4)
- fix(build): stop the extension image tagging a PostgreSQL 17 build as 16 (eb731ea)
- fix(clone): re-measure a table's size before copying it whole (cf2e710)
- fix(export): materialize a lazy clone so the dump is not silently incomplete (0b67714)
- fix(clone): make source-drift detection sound (1a6ed6c)
- fix(k8s): stopped-instance fast-fail uses NotAvailable, not Internal (5ab4fe6)
- fix(db-user): case-fold reserved-role guard + expose apply-preset (63366a8)
- fix(compute-k8s): fast-fail user/exec ops on a stopped instance (F-06) (22ba3f1)
- fix(db-user): close blind-QA gaps F-02, F-03, F-04 (75ad759)
- fix(db-user): adapt to PR 135 (MCP runtime factory + exec-exit-code parser) (ca4cf54)
- fix(db-user): reconcile with upstream-sync API changes (9d01ebd)
- fix(db-user): make apply_preset declarative (revoke-then-grant) (fcda972)
- fix(db-user): scope preset default privileges to deploy owner (856fb01)
- fix(k8s): treat absent exec Status as failure, not false success (b6eb370)
- fix(user): k8s exec exit code + drop-guard + list-hide + password echo (d562922)
- fix(user): reassign owned objects before drop to prevent data loss (21e9f3a)
- fix(user): reject empty passwords + map validation errors to InvalidInput (not Unsupported) (1df6af9)
- fix(user): DROP OWNED BY before DROP ROLE so preset users are droppable (513aa0b)
- fix(mcp): honor GFS_RUNTIME_PROVIDER instead of hardcoding Docker (58e0522)
- fix(compute-k8s): re-apply the custom database user on checkout reprovision (c590328)
- fix(compute-docker): make gfs-postgres init work on podman-machine (macOS) (ec0dd6b)
- fix(cli): honor 'gfs query --database' on the k8s runtime (0fd6f6e)
- fix(compute-k8s): re-apply database name on checkout reprovision (483253d)
- fix(compute-k8s): return the real exec exit code (3f42bea)
- fix(compute-k8s): keep snapshot reclaim in destroy teardown, add preserving teardown for restore (3d4034b)
- fix(clone): fail on version-probe error instead of defaulting to pg16 (f3b6cf2)
- fix(compute-k8s): reclaim snapshots only on destroy, not on checkout restore (241c019)
- fix(cli): persist credentials so Docker checkout re-applies them (complete #1) (b619630)
- fix(port,status,logs): stable host port across checkouts + quieter runtime noise (49b4dee)
- fix(cli,checkout,k8s): honor CLI credentials, atomic failed checkout, reclaim snapshots on destroy (7900aa7)
- fix(tests): drop stale GfsConfig.remote field from test fixtures (89d130b)
- fix(cli): install rustls aws-lc-rs CryptoProvider at startup (k3s TLS) (8a96310)
- fix(commit): don't require a write-freeze on atomic-snapshot runtimes (k8s) (5aad7f9)
- fix(postgres): listen on all interfaces so Service/NodePort is reachable (51894b8)
- fix(compute-k8s): run postgres pod as root so it can init on a fresh PVC (644d5dc)
- fix(k8s): surface task pod log-fetch errors instead of swallowing them (9b227e9)
- fix(k8s): report terminating pod as Stopping, not Running (aeb5c40)
- fix(test): restore display_name in query_in_instance_it fixture (8e4333f)
- fix(k8s): lazy-clone task logs and DNS (3825b8f)
- fix(clone): version bootstrap task image from repo (5e2d05d)
- fix(clone): channel database version into clone default image (0f144ca)
- fix(clone): add lazy-clone detach trait API (ffac1de)
- fix(k8s): lazy clone on split k3s agents (ee24696)
- fix(cli): checkout parity + e2e (de3f8c9)
- fix(schema): capture per-commit schema DDL via sidecar stdout (k8s) (1e2962f)
- fix(k8s): grant secrets RBAC and write credentials via data field (15fc7f0)
- fix(k8s): keep advertised DB credentials truthful across checkout/clone (cadf0f8)
- fix(k8s): use cluster DNS for task sidecars (5e2cde6)
- fix(k8s): return pod env in connection info (ea032c6)
- fix(k8s): auto-allocate NodePort per instance (a6675fb)
- fix(k8s): use hostPort for deploy ports outside NodePort range (c714ed8)
- fix(clone): count real rows when the source size estimate is degenerate (ee8ae15)
- fix(clone): measure real network throughput in cost calibration (7f7b99d)
- fix(clone): make tables with DEFERRABLE unique/exclusion constraints readable (b955c03)
- fix(clone): don't double-count INHERITS child rows on lazy clones (4d66509)
- fix(clone): don't re-fire triggers when hydrating copy-on-read rows (9fe486b)
- fix(clone): carry stateful objects onto lazy clones + fail loud on keyless tables (55f15de)
β‘ Performance
- perf(compute-docker): skip image pull when the image is already local (a497613)
- perf(checkout): skip redundant workspace repair on Docker recreate (c55d447)
- perf(commit): overlap schema-extract with snapshot on k8s (e8b4cd7)
- perf(commit): exponential exec-retry backoff + tighter snapshot poll (9a363cf)
- perf(commit): gate snapshot wait on snapshotHandle, not readyToUse (071bd50)
- perf(commit): capture schema via exec, not a sidecar (c01638a)
π Documentation
- docs(clone): rows_fetched does not count the background worker's copies (3cf4291)
- docs: record a console delete that went through the node daemon and had its stranded volume finished (c48fac4)
- docs: record the runs on a cluster without OpenEBS, on Docker, and reclaiming volumes gfs itself stranded (9be553a)
- docs: reclaim volumes stranded before the fix with gfs storage reclaim, and record the clone-chain run (05040b7)
- docs: record the live k3s runs showing destroy left ZFS volumes behind, and how to reclaim ones stranded before the fix (0408bdb)
- docs: record the live k3s check that a refused checkout destroys nothing (6f57d9d)
- docs(findings): record a forced-VM-stop power-loss analogue for commit (1048cf1)
- docs(findings): record the syscall-level verification of durable writes (96d88e2)
- fix(fsck): an empty HEAD is a hole, and safe_to_remove stops claiming content (acccb42)
- refactor(cli): ask one question about cluster identity instead of three (da75945)
- docs(fsck): document exit 3, the grace floor, and where main left the RFC behind (b16b8f3)
- docs(fsck): the retention setting exists, and ignoring it is a choice with a cost (8b02d18)
- docs(fsck): stop justifying the workspace report with behaviour main removed (910f14f)
- docs(users): name the reserved set where people actually read it (c813fb6)
- docs(rfc): record working copies as a fifth source of garbage (ac8d175)
- docs(rfc): integrity checking and garbage collection (cf8e7d1)
- docs(findings): measure a live Postgres commit on the Kubernetes path (756db6f)
- fix(commit): say that an unfrozen snapshot may be silently lossy (fdf085b)
- docs(findings): record what a commit promises per database engine (44fa428)
- docs(findings): record the live-commit investigation where the code lives (485caad)
- docs(clone): say what the script fix does not cover (a00d7a8)
- docs: stop advertising the repository lock this branch no longer carries (c1d0302)
- docs(skills): correct twelve claims that fail when an agent runs them (615065a)
- docs(commit): pin the drop order the repository lock depends on (579a963)
- docs(skills): say that compute and user do not apply to an embedded provider (83281ec)
- docs: align design-doc titles and tighten wording (7ad2248)
- docs: refine engine module doc comments (ba678a3)
- docs(clone): stop describing a gfs config command that does not exist (8edbeef)
- docs(readme): document what a clone does when the source changes (b5f0061)
- docs(cli/mcp): advertise the user tool + fix Docker-only compute help (d8ebf72)
- docs(clone): correct stale overlay-design docs to the shipped planner-hook (617ca45)
- docs(compute-k8s): note restricted-PSA caveat for run-as-root (harden 3/N) (37d3e94)
- docs: auth smoke and commit known issues (09d1c2e)
β»οΈ Refactoring
- refactor(fsck): name the field for the evidence, not for a conclusion (5f4611e)
- refactor(cli): ask one question about cluster identity instead of three (da75945)
- refactor(k8s-checkout): give the reprovision definition a pure seam (5e53704)
- refactor: move the shared repository lock out to its own change (ace366d)
- refactor(ports): split ContainerProvider out of DatabaseProvider (7627113)
- refactor: extract the database provider catalogue out of compute-docker (21c9c25)
- refactor(providers): collapse the embedded path into a LocalEngine trait (7eefd1c)
- refactor(users): rename the management superuser role (8633187)
- refactor(users): make user management storage-agnostic via a credential port (8b1e5a9)
- refactor(cli): remove the remote runtime mode (61f03ad)
π§ Chores
- chore: bump version to 0.5.0 (128af0c)
- chore(storage-kubernetes): drop two snapshot listers nothing calls (eed66fe)
- chore(sqlite): correct two comments, revive a dead predicate, reject a silently-ignored flag (92aa494)
- chore: remove stale planning and QA scratch docs (325cf2b)
- chore(oss): genericize k8s env + prerequisites (harden 1/N) (43ba646)
- chore(oss): remove remote mode + drop internal-only docs (d76bec3)
- chore(stage0): regenerate Cargo.lock for 0.4.0 workspace (ec9a052)
- chore(lint): clear clippy -D warnings + rustfmt on rebased base (f99f997)
- chore(fmt): apply rustfmt to pre-existing drift (2a7b929)
- chore(ci): clear pre-existing test-compile, clippy, and fmt failures (96e2f81)
- chore(docs): untrack k3s-architecture.md locally (0d26311)
π· CI
- ci(scorecard): re-pin actions/checkout to genuine v7.0.1 SHA (9f7c4de)
- ci: probe whether a reflink works, instead of matching filesystem names (cfed742)
Full Changelog: v0.4.0...v0.5.0