Changelog

0.37.0 - 2026-07-31

Added

• Policy-at-invocation for governed organizations: generated guidance now carries a compact, role-scoped ## Organization Policies consultation index with policy summaries, topics, and digest-verifying my policy show actions. Real AI launches prove every applicable required or optional committed policy blob locally before emitting launch context or starting a harness.
• my compile projects applicable policies as digest-bound policies[] references and fails when a policy's mount is outside the selected role. Non-governed projections and guidance remain unchanged.
• Policy declarations accept optional one-line summary and repeatable topics metadata, including my admin policy add --summary/--topic authoring.
• An approachable governance walkthrough documents the employee my ai experience, agent consultation contract, admin authoring, durable acceptance ledger, and the beta versus experimental boundary.

Changed

• Bare my meetings and my support now list records, including list filters, instead of requiring employees or agents to spell out the default action.
• Active session guidance is regenerated from current organization policy when a session starts, joins a harness, or resumes; harness paths retain governed freshness and digest proof immediately before execution.
• Organization skills keep mutable runtime state under ~/.local/state/my-cli/skills/<manifest>/<install-slug>/. Doctor warns about unexpected skill-source files, and publish/sync holds unadopted files under every declared skill source to prevent credentials or caches entering the manifest.

0.36.0 - 2026-07-21

Fixed

• Automatic publication is target-aware in partial Gnit umbrellas: exact roster members use coordinated publishing, while unrostered My AI checkouts use the guarded built-in path. Session publication uses the same planner, previews run the same local preflight, and only rostered targets can be reported as published by Gnit.
• Coordinated publication now holds before mutation when the roster is invalid, a member identity or control root is unpublishable, a member checkout is missing, or whole-workspace Gnit push would exceed the selected My AI scope.

Changed

• my doctor reports partial Gnit topology, unmanaged or missing members, invalid rosters, and unpublishable control roots without turning supported unrostered content mounts into operator errors.

Added

• An opt-in beta governed-organization foundation: manifest governance declarations, GitHub-backed authorization, digest-bound policy acceptance and launch gates, PR-only governed publication, trusted-base CI validation, and live enforcement audits. Governance completion, record-domain/outbox work, and organization dogfood remain on their active plan and are not claimed complete by this release.
• Experimental access monitoring, recovery capsules, and quarantine machinery behind explicit per-machine activation. Policy and publication flows never activate it; it remains not recommended for real umbrellas until its separate disposable-private-repository revocation drill passes.

0.35.0 - 2026-07-01

Added

• my publish --manifest NAME is the deliberate control-plane publish path: it commits and pushes dirty manifest/catalog control files (manifest.json, catalog/, skills/, guidance/, agent-guidance/), ending the permanently-dirty manifest checkout without routing the control plane through content auto-publish. The low-level form is my sync --publish direct --scope manifest; dirty files outside the control-plane paths still hold.
• my customers add <domain|slug> scaffolds a mounted customers/<id>.md record (--name, --domain, --domain-confirmed, repeatable --alias and --partner); "unknown customer" warnings now point at it instead of keeping a literal slug.
• Held-back my sync results carry a stable reason_code and, where actionable, a next_command (both surfaced by my doctor), including a real first recovery step for dirty-behind and diverged mounts.

Changed

• Handbook mounts default customers/ and fleet/ into their publish paths, and record writes warn at creation time when the new file lands outside the mount's declared publish paths, so records no longer sit unpublished silently.
• Duplicate role/global guidance is de-duplicated in generated guidance and the launch projection.

Fixed

• my sync's active-session hold no longer bounces you toward my session finish --land when the base checkout is dirty (which requireBaseReady would refuse): it now names the dirty base files and tells you to resolve them first.
• Finished or migrated sessions replace their local AGENTS.md/CLAUDE.md with a finished-session stub and surface concrete stale/inactive-session diagnostics, so a shell left inside a landed session no longer dead-ends on stale my work guidance.

0.34.0 - 2026-06-22

Added

• my session start|join|resume|status|list|finish is the primary surface for isolated work units (a git worktree per content mount). my ai --new-session, --session ID, and -r/--resume remain as launch shortcuts; --new-session now prints join/finish hints. Creating a session surfaces its id, path, and mounts plus hints for joining another harness and finishing, and --json carries launch_command, join_command, and finish_command.

Changed

• The worktree work-unit is consolidated onto the noun "session" end to end: on-disk work/<id> becomes sessions/<id>, branches my/work/<id> become my/session/<id>, and default ids are noun-free (YYYY-MM-DD-<hex>). A lazy, idempotent, safe migration moves legacy layouts on session commands and my doctor --fix (never during my ai), skipping ambiguous mounts without mutating them. my work remains as a deprecated alias.

Fixed

• my ai no longer falsely reports workspace guidance as stale (and no longer loops you back to my setup) when the selected role contributes guidance: the launch freshness check is now role-aware, matching my setup and my doctor. Repairable managed guidance (missing, stale, or broken CLAUDE.md alias) is reconciled in place before launch; unmanaged guidance still requires my setup --force.

0.33.0 - 2026-06-17

Added

• Default manifest resolution: with more than one registered manifest, my commands no longer require --manifest. Precedence is explicit --manifest, then the current umbrella's manifest, then a registry default (initially the first-added manifest).
• my manifests default [<name>] [--clear] shows or repoints the global default manifest, and my manifests list marks the active default.

Changed

• Renamed the bundled self-skill from my to my-cli; existing managed installs migrate automatically on the next CLI run or explicit self-skill install, while the canonical id remains my:self.

0.32.1 - 2026-06-16

Fixed

• Release CI: CLI Git test clones now get a local commit identity, so the test suite that goreleaser runs no longer fails on runners without a global Git identity. (v0.32.0 was tagged but failed to publish for this reason.)

Changed

• Docs site: the header and homepage hero now use the text wordmark > my ai instead of an image logo.

0.32.0 - 2026-06-16

Changed

• my sync is pull-only by default. Publishing local changes now requires my sync --push or an explicit --publish mode; manifest sync.publish_policy selects the mode for --push rather than making bare sync publish.
• Human output for setup, sync, and work-session finish is shorter by default, with --verbose available for full row-level detail.
• Agent-operated onboarding now anchors walkthrough commands at the umbrella root and treats explicit "file it" requests as authorization to file a public-safe project issue.
• Work-session guidance now embeds concrete startup context for launched harnesses: umbrella root, organization, selected role, session id/path, mount worktrees, exact resume/finish commands, and the generated base umbrella guidance including organization contract rules. Resuming a session with my ai --session or my ai -r rewrites stale session guidance before launch.

Fixed

• my doctor no longer reports an older cached latest release as if the installed CLI were stale.

0.31.0 - 2026-06-16

Added

• my ai -r / my ai --resume launches a harness in an active work session, with single-session auto-selection, an interactive picker, and deterministic non-interactive errors that list active session ids.

Changed

• Model-driven onboarding now starts a split-pane learn-by-example walkthrough instead of asking for an OK handshake. The operator runs small command sets, the assistant pauses after each set, and onboarding focuses on basic workflows while deeper record and admin work stays agent-operated.
• my work resume is documented as a shell cd helper; use my ai -r [session-id] [harness] to resume work in a harness.

Fixed

• Repo-session documentation now states that catalog code repos launch with my ai --repo and are not landed by my work finish yet.

0.30.1 - 2026-06-16

Changed

• my update now resolves the latest release through GitHub's public releases/latest redirect instead of the rate-limited REST API.

Fixed

• Interactive model-driven onboarding now prompts to replace or skip an existing non-My AI launch skill entry instead of aborting before the harness starts.

0.30.0 - 2026-06-16

Added

• my onboarding --no-agent to force the deterministic walkthrough/status review instead of launching a harness.

Changed

• my onboarding is now the primary guided onboarding command. In an interactive terminal it launches model-driven onboarding by default, auto-detecting a logged-in or singly installed harness when possible; my onboard remains as a compatibility alias.
• Onboarding output now uses readable human prompts and next-step commands instead of tabular status rows.

0.29.0 - 2026-06-16

Added

• my admin roles add|edit|remove and my admin services add|edit|remove for command-driven manifest authoring of role loadouts and service surfaces.
• my onboard --agent [--harness NAME] to launch a harness with the bundled Agent-Operated Onboarding guidance. Zero-manifest runs start the harness from the current directory for AUTHOR bootstrap; registered-manifest runs reuse the normal my ai --setup --no-session launch path for JOIN onboarding.

Changed

• The project is now My AI: the CLI is my, installed from fluxinc/my-cli (https://my-cli.com), with umbrella state under .my-cli/ and MYCLI_* environment variables. Release archives are my-cli_<version>_<os>_<arch>.tar.gz.

0.28.0 - 2026-06-15

Added

• Windows release builds: archives now include windows/amd64 and windows/arm64, so my update can fetch a Windows package.

Fixed

• my update on Windows: extract the my.exe binary from the release archive, and replace the running executable with a rename-aside (Windows locks a running binary, so the old one is moved to <path>.old and cleaned up on a later update).

0.27.0 - 2026-06-14

Added

• Launch-scoped skill composition: my ai composes the organization skills for a launch and materializes them as disposable derived state in the launch root's .agents/skills/ (copied, with .my-cli-managed.json markers and per-slug ownership), instead of installing organization skills into harness user config directories. Non-My AI entries are never clobbered.
• my ai --skills all|none|<id,...> and my ai --profile <id> skill selectors, plus a manifest profiles list of named loadouts. --skills and --profile are mutually exclusive; unsatisfiable skill requirements fail with a precise closure error.
• Harness launch-root skill capability model: Codex and Antigravity read .agents/skills directly; Claude Code gets a generated .claude/skills mirror.
• Antigravity (agy) harness support.

Changed

• my setup, my sync, my manifests sync, and my doctor --fix no longer install organization skills into harness user config directories for launch-root-capable harnesses; those skills are now launch-scoped. my doctor --fix removes leftover user-global organization skills. The bundled my self-skill stays on the global ensure path during migration.
• OpenCode keeps organization skills user-global (it has no launch-root skill discovery); my ai --skills/--profile is rejected for OpenCode.

Removed

• Removed Gemini harness support entirely. Antigravity (agy) is the replacement; my ai gemini and Gemini skill management are no longer supported.

0.26.0 - 2026-06-14

Added

• my onboard: a human walkthrough that explains the model, handles the no-manifest case with registration guidance, offers interactive setup on first run, and records umbrella-local tour completion.
• my setup --interactive: explicit prompt mode for manifest and role selection, including role clearing via none.

Removed

• Removed the deprecated my onboard -> my setup alias. onboard is now a real tour command; setup remains the deterministic machine configurator.

0.25.0 - 2026-06-14

Added

• my compile --role <id> [--manifest NAME] [--home DIR]: emits a deterministic manifest-to-Clawdapus launch projection JSON artifact without launching containers, resolving credentials, or fetching service descriptors.

Removed

• Removed the deprecated my launch dispatch alias. Use my ai for local harness startup and my compile for the contained-runner projection.

0.24.0 - 2026-06-13

Added

• Data bindings may carry domain-notes guidance fragments. Those fragments render into generated AGENTS.md under a labeled, source-attributed ## Domain Notes: <data type> section, kept separate from the organization contract and the my contract verbs.

0.23.0 - 2026-06-13

Added

• Manifest data_bindings: map stable operational data nouns (customers, meetings, support, fleet) to declared mount:<id> or service:<id> surfaces. Mount-backed bindings narrow existing local record commands; service-backed bindings are validated but deferred until service-domain invocation ships.

Changed

• Roles are documented and reported as local loadouts/selections rather than authority grants.

Removed

• Removed the vestigial service grant field from the manifest schema and my services output. Existing manifest JSON that still contains grant remains load-tolerant because unknown fields are ignored.

0.22.0 - 2026-06-13

Changed

• my customers list now reads mounted customers/*.md customer identity records instead of manifest catalog/customers.json, so customer data lives in the workspace data plane and follows the backing Git/API permissions.
• my init scaffolds a workspace customers/ directory instead of an empty manifest customer catalog.

Removed

• Removed my admin customers add|edit and manifest customer catalog validation/loading. Customers are operational records, not manifest admin data.

0.21.0 - 2026-06-12

Added

• my contract list: inspect the organization contract rules in force, with manifest name and the 1-based index used for removal.
• my admin contract add "RULE" --manifest-dir DIR and my admin contract remove <index|"RULE"> --manifest-dir DIR: edit the manifest contract through the standard admin review-commit-push flow, with duplicate, empty, and multiline rules rejected.
• Documentation: new guide pages for Guidance and Contract, Records (meetings/support/fleet), Work Sessions, Services and Roles, and Sync/Doctor/Updates; the site sidebar now covers the full command surface.

0.20.0 - 2026-06-12

Added

• Manifest contract: a list of short, binding organization rules rendered as an ## Organization Contract section in generated AGENTS.md, between the public baseline and manifest guidance fragments. Validation rejects empty, multiline, and duplicate rules; existing guidance drift detection and derived reconcile cover contract changes automatically.
• Added a built-in Fleet Work Contract to generated guidance and the bundled my self-skill: agents should start substantive fleet work with my fleet get, continue a related support record or create a new dated one with my support add, carry fleet identifiers with repeated --identifier, and publish through my sync.
• my fleet get human output now ends with a support-record next step, including an my support add command seeded with the fleet id, known identifiers, and customer when available. JSON output is unchanged.

0.19.0 - 2026-06-12

Changed

• Split the internal/cli implementation into per-domain files without behavior changes: cli.go now holds only the app core, dispatcher, usage, version, and update plumbing, while command implementations live with their domains. The matching CLI tests were split into per-domain _test.go files, leaving cli_test.go for shared helpers and cross-cutting tests.

0.18.0 - 2026-06-12

Added

• Manifest services and roles sections: services describe the organization's remote surfaces (kind: http|mcp, describe_ref, URI auth_ref such as op://, env://, broker://, or none, optional server.json-shaped inline connection for MCP); roles grant mounts, skills, tools, and services. Validation covers ids, kinds, auth/describe references, connection shape, and grant resolution.
• Skills may declare requires: ["service:<id>"] alongside the existing workspace: and tool: forms; my skills show surfaces declared requirements.
• my services list|get and my roles list|get inspection commands with --json.
• my setup --role <id> persists the selected role in .my-cli/state.json, appends role-specific guidance fragments to generated AGENTS.md, and makes the selected role available for role-scoped service materialization.
• my setup (and the derived reconcile in my sync/my manifests sync) now materializes an umbrella-root .mcp.json from MCP services with local connection data (inline connection or a checked-in descriptor), scoped to the selected role. Values pass through as ${VAR} placeholders — never resolved secrets, never network fetches. A hand-written .mcp.json is never overwritten without --force.
• my doctor gained a service section: it reports MCP services without local connection data (URL-only describe_ref), missing checked-in descriptors, unset environment variables referenced by auth_ref or connection placeholders, and op:// references without the op CLI on PATH.

0.17.0 - 2026-06-11

Fixed

• Recording commands (my meetings add, my support add, and my fleet add) now detect when the current directory is inside an active work session and write records to that session's mount worktree instead of silently writing to the base umbrella checkout.

Changed

• my ai now launches from the base umbrella by default. Work sessions are explicit with my ai --new-session or my ai --session <id>; when run from inside an active session, plain my ai <harness> continues launching from that current session. --no-session remains available to ignore a current session for base inspection/admin work.

0.16.0 - 2026-06-11

Fixed

• my ai no longer creates an orphan work session when the requested harness binary is missing from PATH: the binary is resolved before a new default session is created, and the error says no session was created. Resume (--session) and base (--no-session) launches still print the exact fallback command. my ai --print keeps creating a session, as documented.

Added

• my doctor now reports work sessions: each active session shows its live state (ok when clean, warning with dirty/unlanded counts and the matching my work finish command, error when a worktree is missing or git inspection fails), and finished/discarded registry records roll up into a single archived count. The JSON report gains a sessions section.
• Added my work list as an alias for my work status. Human work-session output now includes session-specific follow-up commands: work start prints my work finish <id> ..., and work finish prints a next line for the natural follow-up (my sync, my sync --print, or my work status).

0.15.0 - 2026-06-11

Changed

• Split catalog products from repository checkouts: products in catalog/products.json are now pure business entities (no git_url) that may link implementing repos via repos: ["<repo-id>"], and the organization's repositories live in a new catalog/repos.json inventory. New my repos list|add|remove verbs manage clones under repos/<id> (add is idempotent: an existing clone of the same remote is adopted; conflicting paths hold with remediation). my root/my ai take --repo (--product now errors with the migration hint); my mounts add product:<id> is removed; the sync scope is repos (the products spelling is gone); sync/doctor report repo checkouts with the repo role. Records keep --product as a business reference. Umbrella state migrates automatically (selected_products to selected_repos, product: mount entries to repo:); clones under repos/<id> are untouched. Manifest mount kind repo is removed — repos.json is the single declaration path. A product entry still carrying git_url fails validation with the migration message, so migrate private manifests before installing this release.
• The bundled my self-skill and the docs site now document work sessions fully: the Session concept in the model, the my work start/status/resume/finish verbs in the skill's operational surface and task list, and quickstart/launch guidance reflecting that my ai starts in a fresh session by default.

0.14.0 - 2026-06-11

Added

• Added my work start [--slug] and my work status [--all]: isolated work sessions as visible work/<id>/ directories with a git worktree per synced content mount on a fresh my/work/<id> branch, session-local scratch/, a SESSION.md summary, and a first-class session registry under .my-cli/sessions/. Repo-kind mounts and selected products are not included in session worktrees yet.
• Added my work finish [session-id] --land|--publish|--discard. Landing commits intentional dirty session content, rejects unadopted ?? files and non-content changes, merges clean session branches into the base checkout, removes session worktrees/branches, and records the session outcome.
• my sync (and the targeted sync inside my work finish --publish) now reads the session registry and holds outbound publish of a content mount while any active session on it has dirty files or unlanded commits, naming the session id, path, and the my work finish remediation. Inbound fast-forward pulls are unaffected.
• my ai now launches from a fresh work session by default, supports --session <id> for explicit resume, and requires --no-session for base umbrella or product checkout launches. Added my work resume [session-id] to print a resumable session path.

0.13.2 - 2026-06-10

Changed

• Removed the conflated self-mount compatibility path: mount git_url: "." is now invalid, mounts always materialize as separate umbrella checkouts, and sync emits separate manifest/content entries instead of a merged workspace role.
• Restructured examples/acme-workspace into separate manifest/ and content/ fixture directories.

0.13.1 - 2026-06-10

Added

• Added my record adopt <path> to mark an existing file under a declared content mount as intentional publish content using Git intent-to-add.

Fixed

• Recording commands (my meetings/support/fleet add) now work in a freshly initialized, unpublished organization: local-only mounts count as usable content roots instead of being skipped.

Changed

• my meetings add, my support add, and my fleet add now mark created records with Git intent-to-add so my sync can distinguish My-created records from stray untracked drafts.
• my sync now holds plain untracked (??) files under content paths and names the my record adopt <path> remediation instead of auto-committing arbitrary new files.

0.13.0 - 2026-06-10

Added

• Added my publish: one idempotent command to take a local organization online. It creates private remotes for the content and manifest repositories (or adopts existing origins and pushes, verifying GitHub remotes are private), rewrites local mount URLs to the published remotes in a commit scoped to manifest.json, updates the registry, and prints the teammate join command. --print shows the plan without changing anything.
• Checkouts without an origin remote now report local-only (pointing at my publish) across my manifests sync, my mounts sync, and my sync, instead of failing.
• my sync refuses to publish a manifest whose mounts still reference local paths, and my doctor names each local-path mount with the my publish remediation, so a machine-local URL can never leak to teammates.

Changed

• The manifest is now a control plane separate from workspace content: my init creates two local repositories — a private manifest repo (manifest, catalog, skills, agent guidance) at the registry path, and a content repo at <umbrella>/workspace with the handbook directories. The workspace never contains manifest.json, and hosting permissions can restrict manifest pushes to admins while the whole organization pushes content. Published repos default to <org>-manifest and <org>-workspace. my init --path now selects the content repo location.
• A mount whose git URL matches the manifest's own remote (or the "." marker) remains supported as a compatibility layout: it resolves to the single registered checkout (no duplicate clone, sparse-checkout skipped, one merged sync entry). New organizations get the two-repo layout.

Fixed

• Content publishing in self-hosted organizations is no longer held back by another checkout of the same remote has pending changes: the layouts that kept duplicate checkouts of one remote are gone.
• Gemini skill installs and uninstalls now respect --home isolation instead of writing to the real ~/.gemini (#16).

0.12.0 - 2026-06-10

Added

• Added my init <org-id> to create a small local manifest/handbook repo, commit it, register it, sync the manifest cache, and print the next setup, launch, and optional GitHub publish commands.
• Mount git_url: "." now resolves to the Git URL or local path used to register the manifest, so self-hosted handbook mounts survive publishing.

Fixed

• Reworked the quickstart and manifest docs so the first-run path no longer points at a dead example manifest URL.

0.11.0 - 2026-06-10

Changed

• my doctor now reads as a repair dry run: every finding that my doctor --fix can repair is marked would fast-forward, would reconcile derived guidance and skills, or would reinstall the my self-skill (also would_fix in --json), with a closing fixable count pointing at my doctor --fix. Findings doctor cannot repair keep their explanatory remediation text.

0.10.0 - 2026-06-10

Changed

• Product repositories now clone under repos/<id> instead of products/<id>. my setup migrates an existing products/ directory automatically, and legacy products/<id> checkouts keep resolving until migrated. The sync scope accepts repos (the products spelling still works).
• Startup commands (my root, my ai, my setup) now print a stderr notice line for checkouts the auto-refresh cannot converge — dirty, ahead, behind, or diverged — naming the repository and the command that reconciles it. Stdout is unchanged, so cd "$(my root)" remains safe.

Fixed

• my manifests sync now reconciles generated guidance and manifest skills after pulling or cloning a changed manifest for an existing matching umbrella; pass --no-derived for a cache-only refresh.
• my ai now ensures the bundled my self-skill is installed before it execs a filesystem harness, and manifest skill sync/purge no longer removes that self-skill.

0.9.0 - 2026-06-10

Added

• Added my support list/search/get/add for anonymized support records under support/, with qmd-first search, a built-in markdown search fallback, and linkable frontmatter attribution: an optional canonical customer ID, a repeatable --identifier list for device, order, or asset identifiers, and org member fields (claimed_by, observed_by, and the human sign-off approved_by).
• Added support as a manifest mount kind; handbook mounts without explicit include_paths now treat support/ as approved content for sync publishing.
• Added my fleet list/search/get/add/set and the fleet mount kind: a registry of deployed instances with one record per stable id under fleet/<id>.md, updated in place. my fleet get resolves any entry in a record's identifiers list and reports related support records; my fleet set updates scalar frontmatter while preserving everything else and suggests an my sync --message command for the transition; and my support add warns when an --identifier is unknown to the registry.
• Extracted the shared internal/record engine behind meetings, support, and fleet records (frontmatter parsing now ignores inline YAML comments in unquoted values).

Fixed

• my doctor now reports an absent or stale my self-skill on present harnesses instead of claiming no skill drift, and my doctor --fix reinstalls it (#13).

0.8.0 - 2026-06-09

Changed

• Renamed the CLI from flux to my and the project to My AI (github.com/fluxinc/my-cli); commands read as possessive English.
• flux launch is now my ai; flux onboard is now my setup (deprecated aliases warn on stderr).
• Pluralized noun command groups (manifests, mounts, workspaces); catalog list products is now products list.
• Built-in sync backend renamed from flux to builtin.
• .flux/ is now .my-cli/, ~/.local/share/flux is now ~/.local/share/my-cli, ~/.config/flux/manifests.json is now ~/.config/my-cli/manifests.json, and FLUX_* environment variables are now MYCLI_*.
• Release archives are my-cli_<version>_<os>_<arch>.tar.gz with the my binary inside; the bundled self-skill is my (id my:self).
• my doctor detects legacy Flux state and prints migration remediation.

0.7.0 - 2026-06-09

Added

• Added flux tools list to enumerate manifest-declared tools.
• Added flux admin tools add|edit|remove for manifest tool declarations, with validation and skill-reference checks on removal.

Changed

• Renamed the multi-repo sync backend from Nit to Gnit (--backend gnit, .gnit/roster.yaml, gnit_root in JSON output).

0.6.0 - 2026-06-09

Added

• Added flux update for checksum-verified self-updates from GitHub release tarballs, with --check, --version, --json, and managed-install guidance.
• flux root, flux launch, and flux onboard emit a non-blocking stderr notice when a newer Flux release is available, with --no-update-check, FLUX_NO_UPDATE_CHECK, and FLUX_UPDATE_CHECK_TTL controls.
• flux doctor reports the running Flux version and latest known release.

0.5.0 - 2026-06-08

Added

• flux doctor reports per-checkout Git freshness, derived skill/guidance drift, and the last sync/publish audit.
• Added flux doctor --fix for guarded fast-forward remediation of clean stale manifest/content checkouts plus derived skill/guidance reconcile.
• flux sync records non-print runs to .flux/last-sync.json; doctor fetches refs by default and supports --no-fetch for offline freshness checks.
• flux sync reconciles generated guidance and manifest skills after manifest checkout changes, with --no-derived as an escape hatch.
• flux root, flux launch, and flux onboard perform a best-effort, TTL-gated refresh for clean manifest/content checkouts, with --no-refresh, FLUX_NO_AUTO_REFRESH, and FLUX_REFRESH_TTL controls.
• Manifests can set sync.publish_policy to auto, never, or pr as the default for flux sync when --publish is omitted.
• Added flux admin customers add and flux admin customers edit for customer catalog writes.
• flux admin skills remove reports orphaned tools and allowed skill namespaces, and can remove them with --prune-orphans.

Fixed

• flux meetings add handles date-prefixed slugs and comma-containing attendees/partners correctly.
• Manifest writes omit zero-value JSON fields instead of adding unrelated serialization noise.

0.4.0 - 2026-06-08

Added

• Added the bundled flux CLI self-skill and flux skills self install|uninstall|status.

Changed

• The installer and flux onboard now install the bundled Flux self-skill into existing harnesses, and human CLI runs quietly refresh already-installed file-based copies.
• Documented the public/private skill model: the flux self-skill ships in the public binary; organization skills stay private to a manifest you control.

0.3.0 - 2026-06-08

Added

• Added flux sync: gnit-first bidirectional umbrella reconciliation with a conservative auto policy (direct-push only private, content-only changes).
• Gnit is the multi-repo publish backend once the umbrella is a Gnit control workspace; a guarded Flux Git path is the fallback. Same-remote duplicate checkouts are detected and held when unsafe.

0.2.0 - 2026-06-03

Added

• Added flux skills show and flux skills status.
• Added --skill filtering for skill install, uninstall, sync, purge, and status commands.
• Added flux skills sync and flux skills purge for local harness reconciliation.
• Added flux admin skills add and flux admin skills remove for manifest skill authoring against a maintainer checkout.
• Added admin aliases for mutating/configuration commands.
• Added release packaging, checksum-verified install script, and GitHub Pages documentation site.

Changed

• Clarified that operational reads stay top-level while admin commands mutate shared or workspace configuration.
• Preserved top-level mutating forms as quiet compatibility aliases.

0.1.0 - 2026-05-21

Added

• Added flux customers list for canonical customer IDs, aliases, domains, and partner metadata from manifest customer catalogs.
• Added customer alias resolution for flux meetings list, search, and add.
• Added meeting metadata for partners, attendees, and source IDs.
• Added qmd-first meeting search with built-in markdown search fallback.
• Added configured umbrella discovery for meeting commands run outside the umbrella directory.
• Added flux version and flux --version.

Changed

• Updated generated workspace guidance and the example handbook skill.
• Documented the split between skill materialization and regenerated workspace guidance.