--- title: Changelog description: Release history for kata last_edited: 2026-10-10 --- All notable changes to kata, grouped by release. Versioned releases start with 0.5.0; earlier entries are a retroactive project history grouped by ISO week. ## Unreleased - Walk `kata --help` to find the agent workflow, issue conventions, and command examples. Related commands are grouped, and subcommands list global flags compactly. `kata label remove` aliases `kata label rm`. - Read the daemon's next-step hint in human, agent, and JSON errors. Agent errors put it on a separate `Hint:` line. Hints name CLI flags, such as `--force-new` and `--force`, instead of HTTP request fields. Unknown flags point to the command's help. `close` flag mistakes exit 2 and malformed `--evidence` values exit 3, instead of 1. - Count browser and terminal screen visits with anonymous `screen_viewed` events. Fixed `screen` names count once per installation per UTC day across interfaces and daemon restarts; `surface` identifies the first interface. Daily claims live in exported `meta` rows. `KATA_TELEMETRY_ENABLED=0` opts out. - Anonymous usage reporting includes browser and terminal session duration buckets. Browser time excludes hidden tabs and authentication recovery. - Diagnose setup with `kata doctor`: read-only checks, stable JSON findings, suggested fixes, active hook availability and sampled hook failure counts. Doctor never starts or repairs a daemon; a stopped local daemon is informational. Failed checks use `checks_failed`; reaching the history scan limit is informational. - Manage coding-agent hooks with `kata agent-hook`, matching Roborev and Forge. **Breaking rename:** `kata agent-hooks` now fails as an unknown command; there is no alias. Update scripts and reinstall user hooks with `kata agent-hook install`. For workspace hooks, re-run `kata agent-hook install --local` or the `kata init` hook option that installed them. Reinstalling replaces hooks that still call the plural command. See the [agent-hook reference](reference/cli.md#agent-hook). - Keep OpenClaw inbox context off by default. Reinstalling older managed bundles that predate the opt-in turns sharing off. Enable it with the single-target `kata agent-hook install openclaw --share-inbox` only when every prompt handled by that plugin may receive the Gateway's `KATA_INBOX_USER` inbox. Use `--share-inbox=false` to disable it without removing other hooks. - Set up consumer Muse with standing instructions, a Kata skill, and a recurring poll specification using `kata agent-hook instructions install muse --home `. Preview the files with `--dry --json`. Reinstall preserves file bytes; uninstall keeps foreign content. This relies on instruction-following and provides no native attention support. Choose CLI token-file or connector authentication explicitly. See [Hookless harnesses](workflows/agents.md#hookless-harnesses). - Enroll external agents against login-mode federation hubs and run the returned join command on the spoke. Join reads current hub metadata when it runs, including after a hub purge. **Enrollment credential change:** callers that used `KATA_AUTH_TOKEN` for the hub must now pass `--hub-token-env ` or configure a daemon catalog entry matching `--hub-url`. The local daemon token is no longer sent to the hub. See the [enrollment runbook](operations/federation.md#external-agent-onboarding-without-hooks). - Set up configured coding agents with `kata agent-hook install`, or choose names explicitly. Setup covers 18 native targets and now enables the contract plus every available attention hook by default; scripts needing contract-only setup should pass `--contract-only` or `--attention=false`. Default uninstall removes the full owned bundle. Use `--local` for project scope or `kata init --agent-hooks=codex,pi` during initialization. Project installs use the portable `kata` command unless an executable override is supplied. OpenCode selects its API from a bounded runtime version probe; Muse asks for managed environment permission only during explicit interactive user setup and otherwise reports partial setup. That user-wide allowlist is forwarded to every managed hook, including `KATA_AUTH_TOKEN` when set. Uninstall retains the allowlist; remove names from Muse settings to revoke it. Existing generated commands, custom hooks and the older init flags remain compatible; repeatable `--with-agent-hooks` also selects project bundles. Native session ownership fences delayed cleanup; Hermes uses terminal finalization and Codex init gains SessionEnd. See [target coverage](workflows/agents.md#hook-target-coverage) for runtime limits. - Inspect both hook scopes with bare `kata agent-hook status`. JSON keeps user rows in `harnesses` and project rows in `workspace.harnesses`, with project details under `config`. An unreadable or malformed selected config now exits nonzero and returns the partial report in `error.data`. See [hook status](reference/cli.md#hook-status-and-output). - Upgrade or remove unchanged generated agent extensions after Kata's templates change. Stored digests distinguish installed code from local edits; edited code, metadata and package manifests remain untouched. Reinstall also reuses an owned OpenCode API selection when its runtime probe is unavailable. Hook publication locks now live outside workspace configs, in the OS temporary directory. - Retry hook setup after an init publication failure without losing the project result. Init errors now report whether the project was created and bound, alongside the hook error. All Codex init selectors apply the same user-hook deduplication and tracked-workspace policy. See the [init reference](reference/cli.md#workspace-initialization). - Run `kata agent-contract-hook` directly to emit the built-in Codex briefing. Optional `--source ` replaces the entire prompt with a local UTF-8 file; missing files use the default and empty files supply an empty prompt. Native contract commands share the option. Relative sources stay within the invocation directory; parent traversal and symlink escapes fail. Default hook installations use bare commands, preserve custom prompt hooks, and normalize old generated source markers. The shipped `--source kata-agent-contract-hook` marker still selects the built-in contract; prefix that filename with `./` to select it as a file. Visible and native attention commands accept only the legacy source marker matching `start` or `end` for compatibility; rerun the matching init option to normalize installed hooks to the bare form. Codex user hooks suppress the workspace contract only when both native command platforms are complete and no explicit timeout is below 10 seconds; normalize an older incomplete entry with `kata agent-hook install codex` before rerunning init. - The web UI reports an anonymous `app_opened` event through the daemon when it loads and on the first focus of each later UTC day, so usage reports can count installs where someone opened the UI. `KATA_TELEMETRY_ENABLED=0` turns it off with the daemon heartbeat. Uses Kit v0.31.1. - `kata tui` reports an anonymous `app_opened` event through the daemon once per launch, so usage reports count people who only use the terminal UI. The event carries `surface` (`web` or `tui`) so reports can tell the two apart. `KATA_TELEMETRY_ENABLED=0` on the daemon turns it off. - CLI commands run in a terminal now report an anonymous `app_opened` event with `surface: cli` through the daemon they used, so usage reports count people who only use the command line. Each `KATA_HOME` reports once per UTC day to each daemon, so later commands that day send nothing. Piped output and ordinary agent-mode commands send nothing. `KATA_TELEMETRY_ENABLED=0` on the daemon turns it off. - See daily hook and MCP activity separately from human use. Successful execution hooks, daemon hook children, and admitted MCP calls now send `agent_active` once per UTC day, plus `agent_call_count` at calls 11 and 101. The highest bucket gives observed daily volume. MCP reports use a bounded queue without delaying tools; hook exit reports wait at most 100 ms. Attention hooks and daemon hook children require a daemon target from their operation. Version/help probes without a target send nothing. Reports share the daemon's anonymous install ID and `KATA_TELEMETRY_ENABLED` opt-out. See [telemetry](https://github.com/kenn-io/kata/blob/main/docs/reference/configuration.md#telemetry) for delivery limits and restart behavior. - Use OpenAI-compatible embedding servers that reject a `dimensions` field. Embedding requests no longer send `"dimensions": 0` unless dimensions are requested, so servers no longer refuse the request or return empty vectors. When the embedding server refuses every request, for example because the model name is wrong or it rejects a field, Kata waits the full retry backoff (5 minutes by default) instead of retrying quickly, and leaves documents pending. Project activity cannot bypass this delay; `kata daemon reload` retries immediately after reloading embedding credentials. Uses Kit v0.29.2. - Keep semantic search indexing past an issue that is too long for the embedding model. The embedding server's error now decides the skip: an issue it reports as too long or refused by policy is skipped and the rest keep indexing, instead of every later index pass stopping at that issue. - Hand work between agents over MCP: `kata.load_coordination` adds `kata.assign`, `kata.unassign` with an `expected_owner` guard, `kata.inbox`, and `kata.status`. `kata.project_show` reads one project with its aliases, and `kata.load_docs` adds `kata.search_docs` and `kata.read_doc` over the documentation bundled in the binary. The server now uses MCP Go SDK 1.8.0; clients on older protocol versions keep working. See the [MCP reference](reference/mcp.md#progressive-tool-catalog). - Find every-session contract setup in `kata init --help`, `kata quickstart --help`, the quickstart guide, and the README. Install it with `kata agent-hook install --all`; Codex requires trust through `/hooks`. See [Contract in every session](workflows/agents.md#contract-in-every-session). - Load the contract in every coding-agent session with `kata agent-hook install --all`, remove user hooks with `uninstall`, and inspect user/workspace registrations with daemon-free `status`. Repeated canonical installs preserve file bytes and hook indexes. Codex hooks still require `/hooks` trust; Hermes injects on its first `pre_llm_call` turn. See [Contract in every session](workflows/agents.md#contract-in-every-session). - Run discoverable contract and attention hooks with [`kata agent-hook`](reference/cli.md#agent-hook). Contract responses use each supported harness's native format, including Cursor SessionStart context and Hermes's first `pre_llm_call` turn. Claude Code and Codex can use the visible workspace attention commands. The canonical contract text stays unchanged. - Remove an untracked duplicate Codex contract hook by re-running `kata init --with-codex-hooks` when the selected user config already supplies the contract. Attention hooks keep their positions when unchanged, and init reports any surviving hooks that move and require re-trust. Init installs or updates both hooks in tracked files for teammates without a user hook. See the [init reference](reference/cli.md#workspace-initialization). - Diagnose missing embedding credentials and provider access failures through health and search. Explicit semantic/hybrid searches report readable errors; default search labels its lexical fallback. Keyless local providers continue to work without a placeholder key or an `Authorization` header. - Load embedding keys through `search.embeddings.api_key_file` and refresh them with `kata daemon reload`. Credentials resolve inline > file > environment; a failed selected source blocks provider requests and preserves the backlog. See the [configuration reference](reference/configuration.md#semantic-search) for file requirements. ## 0.18.0 2026-09-17 Kata 0.18.0 helps you hand work to teammates, request their attention, and limit a worker's access to one issue and its children. You can also find work by status or age and receive inbox requests when planning dates arrive. [Release downloads](https://github.com/kenn-io/kata/releases/tag/v0.18.0) · [Changes since 0.17.2](https://github.com/kenn-io/kata/compare/v0.17.2...v0.18.0) **Before upgrading** - Upgrade every participating daemon before relying on federation to preserve teammate attribution. See [teammate workflows](workflows/agents.md#teammate-heads-up). - Upgrade remote daemons alongside the CLI and MCP server. Status-filtered search requires daemon API `0.20.0`; oldest-first lists require `0.21.0`. Kata 0.18.0 includes API `0.21.0`. Release and API versions are separate; see the [HTTP API version history](reference/http-api.md#version-history). - Issue-scoped credentials require [token identity mode](operations/remote-daemon.md#identity-tokens) and must be created on the authoritative daemon, not a spoke replica. **New features** - Request attention with `kata notify --to --message ` and read requests with `kata inbox --for `. Clear a request after handling it. Closing an issue hides its requests; reopening restores any that were not cleared. See [Teammate requests](reference/cli.md#teammate-requests). - Attribute new issues, comments, and attention requests to a teammate: a temporary participant working under an existing actor's identity. Set `KATA_TEAMMATE` or pass `--teammate`, and address a request to `actor/teammate`. The accountable actor and issue owner stay separate from teammate attribution. See [Agent workflows](workflows/agents.md#teammate-heads-up). - Delegate one issue and its current descendants in the same project with an expiring credential through the [CLI](reference/cli.md#remote-and-identity-tokens) or [MCP](reference/mcp.md#events-tokens-and-federation). Workers can read and complete that work and create children under an accessible parent. They cannot access unrelated issues or create more credentials. - Inspect credential scope, expiry, revocation, and last observed use in the TUI and the web app's [Credentials view](guide/web-ui.md#audit-provisioned-credentials). Recorded use is best-effort, not a complete access log. - Find open teammate work grouped by author and teammate in the web app's [Delegated view](guide/web-ui.md#navigate-projects-and-collections). - Receive an inbox request when a schedule or deadline is reached. Kata addresses the owner, or the author if the issue has no owner. Existing requests remain in place until cleared, and acknowledged notices do not repeat for the same date and recipient. See [Planning dates](reference/metadata.md#reserved-keys-vs-opaque-pass-through). - Search only open or closed issues with `kata search --status open` or `--status closed`. HTTP and MCP searches accept the same filter. Omitting it searches both statuses. See [Search](guide/semantic-search.md). - List the oldest matching issues first with `kata list --sort oldest`, including across projects with `--all`. Sorting happens before `--limit`; human output uses a flat list to preserve that order. See the [CLI reference](reference/cli.md#issue-lifecycle). - Connect federated projects using an operator-supplied credential helper instead of copying tokens manually. Choose read-only replication, collaboration, or explicit migration. Status shows approval and any supplied expiry. See [External credential providers](operations/federation.md#external-credential-providers). - Discover running HTTP MCP listeners with `kata mcp status --json`, including ports, backend targets, and token-file paths. Use `kata mcp serve --runtime-dir` to select an already running local daemon. See [MCP listener discovery](reference/mcp.md#discover-running-http-listeners). **Improvements** - Open issues using browser links in show, list, ready, and search results. CLI JSON exposes `web_url`, and MCP issue summaries include links. Daemons without a browser UI omit them. See [Web UI](guide/web-ui.md#open-the-application). - Find the existing issue when creation is refused because of a conflict. CLI JSON includes the error details, and agent output includes matching issue references. - Quit the TUI without confirmation by setting `confirm_quit = false` under `[tui]`. Confirmation remains enabled by default. See [TUI preferences](reference/configuration.md#tui-preferences). **Bug fixes** - Keep the local daemon and web UI current after `kata update`. After replacing the binary, the updater asks a running daemon to restart with its existing settings. A stopped daemon stays stopped. Older daemons require a manual `kata daemon restart`; see [upgrade instructions](get-started/install.md#upgrading-to-0180). - Use Kata's MCP tools with the Anthropic Messages API without tool-schema rejection. Tool inputs no longer use unsupported top-level `oneOf` or `allOf` combinations. - Read issue titles while editing in a narrow web pane. Below 640 pixels, the editor places the title above its actions. - Follow destructive-command confirmation help without a rejected value. The help now shows the required project-qualified issue ID. - Avoid connector readiness failures caused by temporary Windows file-sharing violations. Kata retries those reads. **Acknowledgements** - Thanks to [Rusty Shackleford](https://github.com/salmonumbrella) for teammate attribution and inbox requests, issue-scoped credentials, the Delegated view, planning-date notifications, search status filters, and documentation. - Thanks to [Rod Boev](https://github.com/rodboev) for oldest-first lists, the TUI quit setting and layout work, the Delegated view JSON fix, and Windows and test reliability improvements. - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for browser links, MCP listener discovery, clearer create-conflict errors, and API client improvements. - Thanks to [Ruaridh Williamson](https://github.com/ruaridhw) for readable issue titles in narrow panes and corrected destructive-command help. - Thanks to [Henry Burden](https://github.com/hsb3) for MCP tool schemas that work with the Anthropic Messages API. - Thanks to [Wes McKinney](https://github.com/wesm) for federation credential helpers, daemon restart after updates, and release documentation. ## 0.17.2 2026-09-08 **Bug fixes** - Commands such as `kata tui` no longer fail during daemon upgrades or later startup when the operating system reuses an old daemon's process ID for another process. ## 0.17.1 2026-09-08 **Bug fixes** - Restore the complete release downloads, including Homebrew archives, Debian and RPM packages, and a source archive with the dependencies and web assets needed to build Kata offline. This release contains the same application code as 0.17.0. It repairs release packaging and includes the finalized 0.17.0 documentation. ## 0.17.0 2026-09-08 Kata 0.17.0 helps agents claim work and retry a close when they did not receive its response. CLI commands start faster, simple remote writes need fewer requests, and applications that embed Kata can serve its browser UI below a URL path. **Before upgrading** - Upgrade remote daemons alongside the CLI. The faster remote writes send project names in the write request; older daemons reject that form. - Update custom API clients for empty collections (`[]` and `{}`), case-sensitive request field names, and rejection of `null` arrays. See the [HTTP API contract](reference/http-api.md#version-history). - The SQLite history fix prevents loss during future upgrades. It does not restore history lost during an earlier upgrade. Keep a [backup](operations/backup-restore.md) before upgrading. **New features** - Check issue ownership with [`kata status `](reference/cli.md#issue-lifecycle). It reports the issue's status and revision, your effective identity, the owner, and whether a federation write lease is active, expired, or pending. A lease reserves an issue for a holder while it is live. - Extend a timed federation lease without releasing it first with [`kata federation lease renew --ttl `](operations/federation.md#leases-and-write-gates). Renewal accepts durations from 60 seconds to 24 hours. - Claim an issue only if nobody owns it with `kata claim --if-unowned`. Use `kata unassign --expect-owner ` to remove an assignment only if the owner still matches. See [Claim work](workflows/agents.md#claim-work). - Retry an issue close after a lost response with `kata close --idempotency-key`. The key identifies a request so Kata can recognize a retry. Reusing it for the same request returns the original result and avoids duplicate follow-up comments. Add `--if-match` to reject the close if the issue has changed. See [Close only when verified](workflows/agents.md#close-only-when-verified). - Mount the browser application below a URL path when [embedding Kata](development/embedding.md#mount-below-a-url-path). Navigation, API requests, sessions, live updates, and assets stay under the selected path. - Grant database-backed identity tokens permission to administer connectors and links to external issues with `allow_identity_connector_administration = true` alongside `require_token_identity = true`. This setting defaults to off and grants every active identity token that permission across the daemon. See [Token identity mode](reference/configuration.md#token-identity-mode). **Improvements** - Start CLI commands with less delay. In a 200-run cold-start comparison, median time for a validation command fell from 19.8 ms to 8.9 ms. - Send simple remote `create`, `edit`, `comment`, and `label add` commands with one HTTP request instead of three. Commands involving relationships still require a separate project lookup. See [Remote daemon](operations/remote-daemon.md). - Read more context in `kata search --agent` results, including owner, priority, revision, and a body excerpt. Agent output for `list` and `show` also includes the revision needed for `--if-match` writes. See [Agent output](reference/agent-output.md#reads). - Receive empty JSON arrays and objects instead of `null` collections in API responses. Requests now require case-sensitive field names and reject `null` for arrays. - Follow the new step-by-step [guide](https://katatracker.com/guide/) or browse [reference documentation](https://katatracker.com/docs/). Existing documentation URLs redirect to their new locations. **Bug fixes** - Keep audit history from before an issue move during SQLite upgrades and project-filtered legacy exports. Unexplained missing events stop the upgrade and leave the original database unchanged. - Preserve original issue, comment, and relationship authors when adding an existing project to federation across multiple batches, including resumed transfers. - Preserve relationship creation dates in new federation snapshots and subsequent rebuilds on SQLite and PostgreSQL. Older snapshots without dates retain their existing behavior. - Avoid delaying unrelated PostgreSQL writes by skipping unchanged comments during federation rebuilds. - Report `create_outcome_unknown` when a timeout, cancellation, dropped connection, or incomplete response leaves issue creation uncertain. Check whether the issue exists before retrying, and use `--force-new` only after confirming that no issue was created. - Limit similar-issue checks to the first 500 Unicode code points of the title and body, avoiding oversized queries for long issues. - Keep comment retries working after an issue moves between projects. Comment idempotency keys now belong to individual issues. ## 0.16.0 2026-08-27 Kata 0.16.0 connects issues to work in external systems. You can also record completion evidence for work done outside a repository, import one project without replacing the database, and let auto-started daemons stop when idle. **New features** - Link a Kata issue to a top-level item in an external system through an [external root bridge](reference/cli.md#external-root-bridges). Configure a connector with `[[connector]]` in `/config.toml`, then use `kata connector`, `kata bridge`, or the MCP `kata.load_external_roots` tools to find items, map fields, and bind, pause, resume, reconcile, or unbind them. The external item controls the linked title and body. Incoming comments and lifecycle changes sync by default; sending comments back requires opt-in for each binding. - Build connectors with the versioned `kata.connector.v1` protocol, public Go SDK, and conformance checks usable from any language. Each connector keeps the external service's credentials and API calls inside its own process. - Record `external:` evidence when closing `done` work completed by email, phone, or another channel with no repository artifact. The audit keeps this weaker form of evidence visible. If an evidence type is supplied but disallowed, Kata reports that reason instead of saying evidence is missing. - Import one project's JSONL snapshot into an existing SQLite or PostgreSQL database with `kata import --merge`. Other projects stay unchanged. The import preserves project and issue UIDs, refuses UID collisions, skips cross-project links, and leaves imported federation authority disabled. - Let a local daemon that Kata started automatically stop when idle. This is opt-in. The daemon waits for active requests and finite background work; running stdio and HTTP MCP servers keep it alive automatically. Kata logs the idle exit. An explicit `kata daemon start` replaces an auto-started daemon eligible for idle shutdown with one that stays running. **Improvements** - Limit the total tokens sent to an embedding provider with the optional `model_context_tokens` and `max_batch_tokens` settings. Without them, Kata continues to batch by item count alone. - Reject untrusted plaintext daemon targets before sending a request. An HTTP target outside loopback requires private-network trust or an `allow_insecure` opt-in. - Keep `kata wait` error output in the selected CLI format. Kata resolves the target and output mode once per command. - Simplify how API schemas and deprecated federation `claim` fields are produced. JSON types define their own OpenAPI shapes, and `claim` values come from the corresponding `lease` fields. Published schemas and generated clients are unchanged. - Read disabled web controls more easily in dark mode. The remaining actions use shared Kit buttons and theme settings. - Finish accepted hook jobs during daemon shutdown instead of dropping them. Kata cancels running hooks in time to meet the 25-second shutdown limit. `kata daemon restart` waits up to 30 seconds for the old process to exit. **Bug fixes** - Open local `kata ui` tabs without a token login when the daemon has a static API token. Direct loopback sessions use the existing local browser authority. - Load the initial web snapshot in Firefox. The browser's daemon gateway no longer adds a body to GET and HEAD requests. - Keep the task list usable in narrow windows. Below 700 pixels, the sidebar moves into a drawer, and filters wrap to fit the list pane. **Acknowledgements** - Thanks to [Rusty Shackleford](https://github.com/salmonumbrella) for the external root bridges and connector protocol, external close evidence, per-project merge import, and the daemon credential and web action improvements. - Thanks to [codyw912](https://github.com/codyw912) for idle shutdown of auto-started daemons. - Thanks to [Wes McKinney](https://github.com/wesm) for the Firefox web UI startup coverage. - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for the local web UI access fix and narrow layouts, embedding token budgets, and the Go 1.27 upgrade. ## 0.15.1 2026-08-20 Kata 0.15.1 lets MCP clients connect over HTTP, reports which daemon a client will use, and gives clearer errors when a local daemon cannot start. Remote TUI users must supply the same close evidence as other remote clients. **New features** - Connect MCP clients over Streamable HTTP, the MCP protocol's HTTP transport. The listener requires a bearer token read from an environment variable. Existing project scope, actor attribution, daemon authorization, and tool rules still apply. **Improvements** - Discover the selected daemon with `kata daemon locate`. It reports whether the daemon is local or remote, its transport, URL scheme, and request base URL without exposing credentials. - Explain permission failures in the daemon's runtime directory before trying to start it, including when a restricted environment may be the reason. **Bug fixes** - Apply normal close-message and evidence requirements to remote TUI clients. The TUI exception applies only to the owner's local Unix socket and direct loopback connections without forwarding. - Check daemon runtime-directory permissions consistently through Kit's shared validation, including symlink, ownership, and private-directory failures. **Acknowledgements** - Thanks to [Rusty Shackleford](https://github.com/salmonumbrella) for the Streamable HTTP MCP transport, support-aware daemon discovery, and remote TUI close safeguard. - Thanks to [codyw912](https://github.com/codyw912) for clear daemon autostart errors in restricted environments. - Thanks to [Wes McKinney](https://github.com/wesm) for the shared runtime-store writability check and the release documentation. ## 0.15.0 2026-08-16 Kata 0.15.0 lets agents manage more of their work through Model Context Protocol (MCP) tools. It adds scheduling, deadlines, and agent guidance that refreshes during a session, plus clearer errors for local daemon connections. **New features** - Manage issues, projects, activity, recurrences, federation, synchronization, imports, tokens, and storage through MCP. Thirteen section loaders make up to 55 typed tools available as needed, so an agent does not have to load the entire catalog at startup. Tools remain limited to the server's project scope. - Set planning dates with `kata schedule` and `kata deadline`, or the `kata.set_schedule` and `kata.set_deadline` MCP tools. A future schedule keeps an issue out of ready work until that date; a deadline does not. - Read the agent briefing with `kata quickstart --format contract`. It prints the same instructions as `kata init --with-agents`, without file markers or file changes. `kata init --with-codex-hooks` supplies those instructions on session startup, resume, clear, and compaction. - Check the version with `kata --version`. It supports the same human, JSON, and agent output as `kata version`. **Improvements** - Allow an embedded local browser to create a direct-loopback web session when `Origin` is missing or empty. This exception applies only to session creation. Exact Host, loopback, forwarding-header, and cross-site Fetch Metadata checks still apply. - Move older federation bindings to their validated HTTPS endpoint when their `allow_insecure` opt-in was saved only with the same-endpoint credential. These bindings predate storing that setting on the binding itself. - Speed up tests by skipping durability flushes in both SQLite stores when the test-only `KATA_TEST_FAST_SQLITE` mode is enabled. Production durability is unchanged. **Bug fixes** - Report the recorded process ID, endpoint, and connection error when a running local daemon cannot be reached. Kata no longer treats it as stopped and tries to start another daemon. **Acknowledgements** - Thanks to [Rusty Shackleford](https://github.com/salmonumbrella) for the expanded MCP server, planning-date workflows, agent contract output, and conventional version flag. - Thanks to [Wes McKinney](https://github.com/wesm) for originless embedded local sessions, legacy federation rebinding, and the release documentation. - Thanks to [Phillip Cloud](https://github.com/cpcloud) for removing durability flushes from SQLite fast mode. - Thanks to [codyw912](https://github.com/codyw912) for clear unreachable local-daemon errors. - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for the Go 1.26.6 toolchain security update included in this release. ## 0.14.3 2026-08-10 **New features** - Let Forge reuse Kata's issue details, including properties, checklists, links, and comments, through a network-free shared presentation package and bounded daemon reads. **Improvements** - Ensured release artifacts use the requested release tag throughout the automatic and manual publishing paths. **Bug fixes** - Restored login discovery for Web UI tabs opened directly, preserving the server-advertised authentication handoff. ## 0.14.2 2026-08-10 **New features** - Added the official `kenn-io/tap/kata` Homebrew installation path for macOS and Linux. - Added import support for SQLite-era Beads databases through the existing `kata import --source-format beads` workflow. - Added a build-time distribution marker so Homebrew, `.deb`, `.rpm`, and third-party packages retain ownership of their installed binary while `kata update --check` remains available. - Added a checksummed release source archive with production browser assets and vendored Go dependencies for reproducible, network-free Homebrew Core builds. **Improvements** - Upgraded `go.kenn.io/kit` from v0.14.0 to v0.19.1. - Added prerelease-safe publishing so release candidates do not replace the latest stable GitHub release or update the stable Homebrew tap formula. - Kept stable release notes based on the previous stable tag after publishing a release candidate from the same commit. **Bug fixes** - Protected edited TUI new-issue and new-child-issue drafts with discard confirmation when Esc is pressed, without prompting for untouched forms. - Corrected `scheduled_on` timezone handling and civil-time semantics across readiness queries and browser snapshots. ## 0.14.1 2026-08-08 kata 0.14.1 adds a first-class browser workspace and native MCP integration, expands cross-project and label-filtered discovery, and makes release, federation, and daemon operations safer and more observable. **New features** - Added a daemon-served Web UI for managing projects, issue collections, fields, comments, checklists, relationships, recurrences, and multiple configured daemons. - Added a native, project-bound stdio MCP server with 13 structured issue tools, protocol negotiation, fixed actor attribution, and bounded results. - Added `kata list --all`, with status, priority, owner, label, exclusion, and metadata filters that compose across all non-archived projects. - Added repeatable `--label` and `--no-label` filters to project-scoped `kata search`, applied before lexical limits and during vector-result hydration. - Allowed `kata tui ` to resolve and open an issue's detail view directly. - Exposed the standard Go profiling handlers under `/debug/pprof/` on the daemon's existing authenticated listener. **Improvements** - Centralized Claude Code and Codex hook configuration on kit's shared agent-hook manager while preserving the existing init flags, lifecycle matchers, attention behavior, unrelated configuration, and workspace symlink boundary. - Added field-scoped Markdown rendering to `kata show --render` for issue descriptions and comments while preserving literal record structure and machine-readable output. - Made `ready --all` and `next --all` compose with owner and label filters. - Excluded issues parked with `someday=true` or a future `scheduled_on` date from scoped, global, and browser ready queues. - Limited workspace binding discovery to the current Git root, preventing an ancestor repository's `.kata.toml` from capturing a nested repository. - Avoided federation-wide link reconciliation for batches that cannot affect links and gave synchronization requests their own bounded timeout. - Kept the release installers compatible with releases that predate embedded Web UI validation while requiring current releases to validate their assets. - Added daemon API-version preflights before filtered global list/ready and filtered search operations, so older daemons fail with an upgrade message instead of silently ignoring filters. **Bug fixes** - Kept GoReleaser validation from dirtying the release checkout before archive publication, allowing the release workflow to validate and publish the embedded Web UI artifacts successfully. **Acknowledgements** - Thanks to [Rusty Shackleford](https://github.com/salmonumbrella) for the native MCP server, cross-project list filters and version checks, scoped search label filters, parked-ready semantics, and direct TUI issue opening. - Thanks to [Wes McKinney](https://github.com/wesm) for the Web UI, bounded federation work, Git-root workspace discovery, legacy installer support, and clean GoReleaser validation. - Thanks to [Matthew Jacobs](https://github.com/mjacobs) for Markdown rendering in `kata show`. - Thanks to [Joi Ito](https://github.com/Joi) for composing `ready --all` and `next --all` with scoped filters. - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for the standard daemon profiling endpoints. ## 0.13.0 2026-07-31 kata 0.13.0 makes federation endpoint changes safer, automates enrollment, and improves agent and embedded-service integrations. **New features** - Added explicit, retry-safe `kata federation rebind` endpoint migration for moving one spoke or every local spoke to a named HTTPS hub catalog entry. The daemon validates the unchanged hub project identity with the existing enrollment and drains old-endpoint sync work before preserving tokens, capabilities, actors, and sync cursors across the URL/security-only update. - Added startup-configured spoke-to-hub project mappings that create or adopt local projects, enroll them idempotently, retry hub failures without affecting daemon readiness, and expose sanitized aggregate progress in `/health`. Federation teardown remains an explicit `kata federation leave`. - Added project-scoped federation enrollment creation, history, and revocation methods to the mountable Go service for hosts that use the restricted embedding profile. - Added `kata init --with-codex-hooks` to install the `work.attention` harness into a Codex CLI workspace's `.codex/hooks.json`. Only the `SessionStart` half is wired, since Codex has no stable session-end hook event yet; cover the end half with a launcher wrapper until upstream ships one. **Improvements** - Kept remote TUI project resolution path-free by using only portable project names and Git identities for configured remote daemons. - Authenticated federation ingest before body decoding and added bounded host admission for large uploads, returning a retryable `429` response when an embedding host is saturated. - Added federation credential host validation and optional host-side access revalidation so credentials remain bound to their configured origin and embedded hosts can deny revoked outside authority without replacing Kata authentication. **Acknowledgements** - Thanks to [Wes McKinney](https://github.com/wesm) for federation endpoint rebinding, configuration-driven enrollment, embedded federation lifecycle controls, path-free remote TUI resolution, safer federation ingest, and credential host validation. - Thanks to [Matthew Jacobs](https://github.com/mjacobs) for the Codex CLI attention hooks. ## 0.12.1 2026-07-21 kata 0.12.1 hardens daemon credential routing, embedded-service boundaries, and repeatable query filters. **Improvements** - Added restricted embedding policies and transaction fences so host-owned authorization is enforced at the same boundary as Kata writes and worker operations. - Preserved repeated values in query filters, including label, exclusion, metadata, actor, and digest filters. **Bug fixes** - Refused Git-tracked `.kata.local.toml` files so a committed daemon redirect cannot route a global bearer token to an untrusted origin. **Acknowledgements** - Thanks to [Matthew Jacobs](https://github.com/mjacobs) for the tracked-local configuration guard that prevents daemon credential misrouting. - Thanks to [Wes McKinney](https://github.com/wesm) for restricted embedding policies, transaction fences, and repeated query-filter handling. ## 0.12.0 2026-07-20 kata 0.12.0 gives embedded integrations host-controlled access policies and project lifecycle APIs, and makes TUI input safer and easier to navigate. **New features** - Added host-controlled authentication and per-operation authorization for mounted Go services through `Config.Access`. Embedding hosts supply the authenticated principal; kata exposes the matched operation and project scope to their controller, records the host-supplied actor, hides denied resources, and revalidates access during long-lived event streams. - Added retry-safe `Service.EnsureProject` and `Service.ArchiveProject` methods so embedding hosts can provision exact project UID/name bindings and archive projects without deleting their stable identity, task history, or events. **Improvements** - Made Ctrl-O the canonical TUI save/apply chord while retaining Ctrl-S as a compatibility alias for existing users. - Allowed Up and Down to navigate live search results before committing the filter, with reversible query/results focus and an explicit keep-filter exit. - Protected non-empty TUI comment drafts with discard confirmation when Esc is pressed; canceling preserves the full in-memory form state, and the modal footer advertises the active discard and keep-editing actions. **Acknowledgements** - Thanks to [Wes McKinney](https://github.com/wesm) for host-controlled embedded access, project lifecycle APIs, and safer TUI input and navigation. ## 0.11.1 2026-07-19 kata 0.11.1 makes pgvector optional for PostgreSQL deployments that do not use semantic search. **Improvements** - Allowed core PostgreSQL task, federation, and lexical-search storage to start without pgvector. Kata creates vector tables only when pgvector 0.7 or later is already installed in `public`; configuring semantic search without it reports the feature as unavailable without affecting core storage. **Acknowledgements** - Thanks to [Wes McKinney](https://github.com/wesm) for optional pgvector support in PostgreSQL deployments. ## 0.11.0 2026-07-18 kata 0.11.0 completes PostgreSQL support across kata's storage and sharing workflows and exposes the application as a listener-free Go service. It also strengthens machine-readable identity and semantic-search input validation. **New features** - Added complete PostgreSQL support for the storage contract, daemon operation, federation, claims, external and JSONL import, export and atomic snapshot restore, lexical search, and pgvector-backed semantic search. Configure it with `KATA_DSN`, `[storage].dsn`, or a `postgres://` / `postgresql://` URL. - Added `kata storage postgres migrate` and `status`, validation-only runtime configuration, server-identity-verified TLS requirements, and separate schema-owner and serving roles for production PostgreSQL deployments. - Exposed the module root at `go.kenn.io/kata` as a mountable Go service. Applications construct a service with `kata.New`, mount its HTTP `Handler`, run federation, GitHub sync, and timed-claim workers with `Run`, and release owned storage and workers with `Close`; kata does not take over the listener, signals, or host process lifecycle. **Improvements** - Added the stable canonical identity `"name": "kata"` to `kata version --json`, alongside the existing version and build fields, so automation can identify the binary without parsing human-readable output or contacting a daemon. - Rejected embedding responses with null components or zero-norm vectors before they can be persisted or used for a query. Background fills remain pending and retry after an invalid provider response instead of marking unusable vectors complete. **Bug fixes** - Honored `KATA_HTTP_TIMEOUT` for probes of explicitly configured remote servers, while keeping the separate one-second local runtime-discovery probe. Slow healthy WAN and reverse-proxied daemons now get the same request budget as the command that follows the probe. **Acknowledgements** - Thanks to [Wes McKinney](https://github.com/wesm) for complete PostgreSQL support, the mountable Go service, and configured-remote timeout handling. - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for stable JSON version identity and invalid embedding-vector rejection. ## 0.10.0 2026-07-11 kata 0.10.0 makes ready-work selection priority-aware, carries cross-project relationships through federation, and improves daemon and agent-workflow operations. **New features** - Added `kata next`, which deterministically selects one ready issue. Explicit priorities beat unprioritized work, lower numeric priorities win, and ties retain the ready API's order. It supports the scoped `ready` filters, cross-project `--all` selection, compact human/agent/JSON output, and `--full` issue detail. - Rendered parent/child trees in human `kata list` output with box-drawing connectors while preserving server order in JSON and agent output. Children whose parent is outside the fetched result remain visible at the top level. - Added `kata daemon restart`. It validates replacement settings before stopping the current local daemon, waits for graceful shutdown, and starts a replacement with configured or explicitly repeated listener settings. - Synchronized cross-project links across federated projects. Link events are retained when a peer has not arrived yet, then materialize after both endpoint projects join the same hub federation group regardless of project enrollment or synchronization order. - Added `kata federation quarantine list` and `show` so operators can inspect project ownership, event ranges and UIDs, timestamps, and retained errors before retrying or skipping. Federation project detail in the TUI now shows the same retained quarantine errors. **Improvements** - Reported live semantic-search backfill progress through the `/health` `embeddings` object, including start and last-progress timestamps plus a smoothed processing rate and ETA once enough progress samples exist. - Improved human `kata daemon status` output with the daemon address, PID, binary version, and uptime. JSON status includes the database path and start time for programmatic diagnostics. - Added `kata init --with-hooks` for Claude Code workspaces and moved the attention lifecycle logic into the installed `kata` binary. The generated exec-form hooks no longer depend on a repository script whose contents could change behind an already approved command. - Extended the managed block written by `kata init --with-agents` with the `work.branch`, `work.attention`, and `work.attention_msg` conventions. Re-running the command refreshes an older managed block in place. - Improved `kata wait --timeout` and `--poll-interval` validation: a bare number remains rejected as ambiguous, but the error now suggests the seconds-qualified spelling and lists supported duration units. **Bug fixes** - Fixed a federation deadlock where two projects whose first pending batches referenced each other's new issues could both become permanently quarantined. Compatible spokes also resend older push quarantines created by the former missing-peer validator without advancing the cursor; unrelated validation failures remain quarantined. - Preserved labels in project-scoped and cross-project `kata ready` results by returning the same hydrated issue projection used elsewhere in the API. - Fixed GitHub issue and comment pagination when a `Link` header uses GitHub's numeric `/repositories/{id}/...` URL form. kata rewrites that form to the bound owner/repository path before applying its credential egress guard. **Acknowledgements** - Thanks to [Matthew Jacobs](https://github.com/mjacobs) for parent/child list rendering, ready-result labels, duration guidance, `work.*` managed guidance, and the binary-backed attention hooks. - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for federated cross-project link convergence, quarantine recovery and discovery, and live embedding progress reporting. - Thanks to [Wes McKinney](https://github.com/wesm) for priority-aware `kata next`, daemon restart, and daemon status improvements. - Thanks to [Barret Schloerke](https://github.com/schloerke) for numeric GitHub pagination URL support. ## 0.9.0 2026-07-09 kata 0.9.0 adds coordination primitives for agent launchers and dashboards, improves scriptable metadata handling, and refreshes semantic search storage and terminal rendering. **New features** - Added `kata wait`, a read-only fan-out/join command for scripts that need to block until one or more issues close or report attention through `work.attention`. It supports `--until closed|attention|needs-human|stuck`, `--any`/`--all`, polling control, timeouts with a dedicated exit code, and JSON output for wrappers. - Added first-class issue metadata commands and API support. `kata meta set`, `kata meta unset`, and `kata meta get` read and patch issue metadata, support raw JSON values where needed, and expose optimistic concurrency through `--if-match` / `If-Match` revisions. - Added documented `work.*` metadata conventions for branch orchestration workflows, including `work.branch`, `work.attention`, and `work.attention_msg`, plus an agent-orchestration runbook for launchers, working agents, coordinators, and merge automation. **Improvements** - Polished human-readable `kata list` and `kata ready` output with clearer status glyphs, priority chips, label chips, owner display, and summary footers while preserving machine-readable `--json` and `--agent` output. - Improved daemon HTTP API performance by gzip-compressing eligible JSON responses when clients send `Accept-Encoding: gzip`; server-sent event streams stay uncompressed so they keep streaming normally. - Moved semantic search storage from a single embeddings table in `kata.db` to a sidecar vector index built on the shared `kit` vector layer, named after the database file (`kata.vectors.db` for the default `kata.db`), with chunked embeddings instead of a fixed truncation cap and generation-based model swaps: changing `model`, `dims`, or `fingerprint_salt` fills a new generation in the background and cuts over automatically. During that backfill the vector leg is unavailable — `auto` searches degrade to labeled lexical results and explicit `semantic`/`hybrid` requests return 503 until the cutover — instead of losing the vector index outright; lexical search is unaffected. The sidecar is disposable derived state — safe to delete, excluded from backups, rebuilt by re-embedding. - The first daemon start after upgrading re-embeds every issue; the rebuilt index activates immediately and serves partial semantic results while the backfill drains (the `embeddings` backlog in `/health` reports progress). JSONL export no longer carries `issue_embedding` records; import of older archives that still contain them skips those records instead of failing. - Soft-deleting an issue now removes its vectors at the next reconcile, so deleted content is never re-sent to the embedding endpoint by later index rebuilds. Searches with `include_deleted` rank soft-deleted issues lexically only; restoring an issue re-embeds it and semantic recall resumes. - Updated the TUI stack to Charm's v2 Bubble Tea, Bubbles, Lip Gloss, and Glamour packages for improved terminal rendering and input behavior. **Acknowledgements** - Thanks to [Matthew Jacobs](https://github.com/mjacobs) for `kata wait`, issue metadata commands, `work.*` orchestration conventions, and human-readable `list` / `ready` polish. - Thanks to [Wes McKinney](https://github.com/wesm) for gzip API compression, semantic search vector storage, and the TUI stack update. ## 0.8.0 2026-07-04 kata 0.8.0 adds a graph-shaped issue API for clients that need to visualize or analyze connected work, makes daemon projects easier to create without a local workspace, and records testing guidance for behavior-based validation. **New features** - Added `GET /api/v1/projects/{project_id}/issues/{ref}/graph` for reachable issue graphs. The endpoint walks parent, `blocks`, and related relationships from a source issue, supports `depth=full` or a bounded hop count, can hide closed non-source issues with `hide_done=true`, includes cross-project `qualified_id` values, and reports unresolved link endpoints without dropping the rest of the graph. - Added `kata projects create ` for creating or returning an active daemon project by name without writing `.kata.toml`, `.gitignore`, or agent guidance files and without attaching a workspace alias. - Added testing guidance that discourages tautological content assertions, especially shell tests that grep scripts or config for implementation text, and favors assertions against observable behavior, persisted state, command output, API responses, events, or rendered UI. **Improvements** - Documented the name-only project creation workflow in quickstart and CLI reference docs so projects that are not tied one-to-one to a repository can be created before any workspace is initialized. **Acknowledgements** - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for the reachable issue graph API and testing-without-tautologies guidance. - Thanks to [Wes McKinney](https://github.com/wesm) for name-only project creation. ## 0.7.0 2026-06-29 kata 0.7.0 improves discovery, sharing, and release operations. Search can now use opt-in embeddings for semantic recall while preserving lexical behavior by default, GitHub sync gains safer daemon-side credentials and parent-link reconciliation, and federation handles larger adoption and push workloads more reliably. **New features** - Added opt-in semantic search for SQLite-backed projects. Configure an OpenAI-compatible `/embeddings` endpoint under `[search.embeddings]` to make default `kata search` run hybrid lexical/vector search; use `--lexical`, `--hybrid`, or `--semantic` to force a mode. Search remains lexical with no embedding config, and automatic fallback reports degraded mode instead of silently hiding vector failures. - Added daemon-scoped GitHub sync credentials. Shared daemons can use `[[github_sync.app]]` GitHub App credentials matched by `(host, owner)`, env-token fallback is host-scoped with `[github_sync].token_host`, and GitHub Enterprise hosts must be explicitly allow-listed. - Added GitHub sub-issue parent-link synchronization. GitHub-sourced parent links are imported and reconciled as source-managed kata parent links, while unsupported Enterprise schemas preserve existing source-managed parent links instead of deleting them. - Added `kata projects rewrite-author` and the matching HTTP action for project-scoped current-state identity hygiene. It rewrites exact matches in issue authors, issue owners, comment authors, and link authors before export, sharing, or federation enrollment. - Added `kata comment edit` and the matching HTTP route. Comment edits preserve the comment UID, author, creation time, and thread position, which makes them useful for pre-federation content redaction. - Added `kata projects purge` for permanently deleting archived projects and freeing their names, with force/confirmation guards, audit tombstones, JSON output, and federation-binding refusal. **Improvements** - Improved federation adoption by chunking baseline snapshot pushes so large existing projects can be adopted without sending one oversized push request. - Improved federation push reliability by splitting oversized outbound batches and retrying them as smaller batches instead of quarantining only because a request was too large. - Improved `kata update` behavior for development builds. Update checks fetch fresh release data, show the latest official release and artifact metadata, and require `--force` before replacing a dev build. - Published release artifacts automatically from tag pushes through the release workflow. - Served nav-listed Markdown documentation sources from docs builds, so public `.md` URLs come from the same deployment as the rendered pages. - Refreshed release documentation and docs navigation for semantic search, GitHub sync credentials, project purge, author rewrite, and comment redaction. **Acknowledgements** - Thanks to [andy-vdg](https://github.com/andy-vdg) for the scoped GitHub sync service credential work and GitHub parent-link synchronization. ## 0.6.0 2026-06-24 kata 0.6.0 expands kata's release and sharing surface: GitHub Issues can now feed a kata project, private-network daemon deployments have an explicit tokenless write mode for trusted single-user networks, and Windows users have a hosted release installer. **New features** - Added one-way GitHub issue sync with `kata sync github`, backed by daemon-owned bindings, cursors, import mappings, status, and a poller. The first provider imports GitHub issues and issue comments through daemon-side GitHub credentials, skips pull requests, prefixes imported titles by default, and keeps GitHub as the source of truth for synced fields. - Added provider-neutral issue-sync API/storage foundations so GitHub sync state participates in backup, restore, JSONL cutover, and daemon status flows. Restored sync bindings come back disabled until explicitly re-enabled on the new host. - Added support for running GitHub sync on federation hubs, so GitHub-origin kata events can replicate to spokes while direct GitHub sync on spokes stays rejected. - Added explicitly enabled tokenless writes and event streams for daemons bound to literal private IP addresses. Operators can opt in with `[auth].allow_unauthenticated_private_network_writes = true` or `KATA_ALLOW_UNAUTHENTICATED_PRIVATE_NETWORK_WRITES=1`; token administration remains blocked without authentication. - Added a hosted Windows PowerShell release installer at `https://katatracker.com/install.ps1`, including release-asset selection, checksum verification, user-local install, and user `Path` updates. **Improvements** - Improved local-first federation resilience by preserving relationship writes to newly created local issues while their create or snapshot events are still pending push to the hub. Once the hub has acknowledged the materializing event, later missing-issue responses remain visible errors. - Tightened the pending-push federation exception to `issue_not_found` errors so broader hub route or project misconfiguration is not hidden. - Changed explicit `kata daemon start` to start a background daemon by default and return after startup is confirmed; `kata daemon start --foreground` remains the service-manager and hosted-deployment mode. - Sped up Windows development and release validation by moving broad CLI and daemon handler fixtures off slower git/init paths when tests only need a seeded project. - Revamped the README and docs front page into a clearer landing page with direct install, quickstart, and feature-orientation paths. - Expanded the 0.5.0 changelog into a fuller first-release history. **Bug fixes** - Restored TUI daemon selection recovery when switching from the daemon selector to a daemon with no registered projects. Escape now returns to the selector for that switch path while direct empty-daemon startup still shows the onboarding state. - Preserved the federation trust boundary for delayed push scenarios without hiding real broken mappings or missing hub state after the pending local issue has materialized. ## 0.5.0 2026-06-22 kata 0.5.0 is the first versioned public preview release. It includes the core local-first issue tracker, a full terminal UI, agent-oriented workflows, hub-and-spoke federation, portable backup/import paths, and the first release automation for binary distribution. **New features** - Local-first issue tracker backed by a daemon-owned SQLite database, workspace bindings, project discovery, project aliases, stable issue ULIDs, and short refs derived from those ULIDs. - Issue lifecycle commands for creating, listing, showing, editing, commenting, closing, reopening, deleting, restoring, and purging issues. - Labels, ownership, assignment, claim/unclaim workflows, parent/child hierarchy, blockers, related links, cross-project links, and cross-project ready views. - BM25-ranked issue search, look-alike duplicate protection, create idempotency keys, and safe retry behavior for automation. - Interactive `kata tui` with project switching, nested and flat issue lists, issue detail pages, filters, search, inline issue creation, editor-backed body/comment editing, mutations, help, split-pane layout, and realtime event refresh. - Durable event polling and SSE streams, including `kata events --tail` for NDJSON consumers and reset handling after destructive history changes. - Daemon hooks with TOML configuration, bounded queues, worker pools, timeouts, output capture, log rotation, pruning, and reload support. - JSONL export/import for backup, migration, schema cutovers, and project restore workflows, plus importer support for Beads projects. - Hub-and-spoke federation for shared kata projects, including enrollment, actor attribution, shared daemon catalogs, TUI daemon switching, and reversible spoke leave/rejoin. - Remote-client mode with trusted private-network bearer auth, token-based identity, trusted-proxy actor headers, and explicit daemon selection. - Hosted-mode daemon support for platforms that provide a `$PORT` contract, including Cloud Run, Render, Fly.io, Railway, App Engine, and similar hosts. - Windows daemon support alongside Unix socket operation, with runtime addresses published as URLs for clients. - OpenAPI support with `kata openapi`, a committed schema, generated client artifacts, and API schema version reporting in `/health`. - Release tooling for annotated tags, tag verification, GitHub release artifacts, self-update discovery through `kata update --check`, and installable release packages. **Improvements** - CLI ergonomics now include `kata version`, `kata whoami`, `kata health`, project list/show/rename/merge/restore commands, daemon start/status/stop/log/reload commands, and `--comment` support on mutation commands. - Agent-facing output gained a stable text format, better close-justification safeguards, claim-oriented ready filters, and `kata init --with-agents` for writing kata guidance into `AGENTS.md` or `CLAUDE.md`. - TUI rendering gained hierarchy controls, labels in list/detail views, responsive narrow layouts, strict status filters, ANSI-safe text handling, no-color support, bracketed paste, opt-in mouse support, and cleaner reconnect behavior. - Documentation now covers installation, quickstart workflows, agent usage, configuration, remote daemons, federation, hosted deployment, backup/restore, OpenAPI compatibility, SaaS issue tracker comparisons, design notes, and docs-site maintenance. - Release builds now publish archives, checksums, Linux packages, Windows binaries, and metadata names that match the installer and updater discovery contract. **Bug fixes** - Fixed active daemon selection, stale daemon upgrades, remote-client resolution, named daemon auth isolation, and schema-skew quarantine recovery. - Hardened credential routing so local daemon tokens, catalog daemon tokens, and explicit `--hub-token` paths do not cross unintended origins. - Hardened JSONL import/export, destructive command confirmation, purge audit records, schema migration paths, and orphaned-state recovery. - Improved SQLite contention handling with retries for transient write locks. - Fixed percent-encoded git remote URL handling during init. - Fixed optional OpenAPI client response modeling and generated-artifact drift checks. - Relaxed close throttling so safeguards remain useful without blocking legitimate sibling issue closures. **Acknowledgements** - Thanks to [Marius van Niekerk](https://github.com/mariusvniekerk) for TUI navigation and presentation work, Beads import and priority/schema-cutover support, SQLite contention retries, generated OpenAPI client support, issue project moves, daemon lifecycle/git helper adoption, telemetry, and module path migration. - Thanks to [Phillip Cloud](https://github.com/cpcloud) for agent guidance consolidation, hosted-mode `$PORT` binding and docs, trusted-proxy actor headers, active-daemon client handling, Unix runtime URL publication, and the storage abstraction/Postgres schema shell. - Thanks to [Matthew Jacobs](https://github.com/mjacobs) for `kata init --with-agents`, `kata openapi`, the committed OpenAPI schema, `/health` API schema version reporting, and API compatibility documentation. - Thanks to [Andy Hadjigeorgiou](https://github.com/andyxhadji) for ready filters, claim workflows, and the cross-project `kata ready --all` view. - Thanks to [Nat Torkington](https://github.com/njt) for Windows daemon support and percent-encoded git remote URL handling during init. - Thanks to [Jesse Vincent](https://github.com/obra) for consolidating relationship editing into create/edit flags and adding `--comment` support to mutation commands. - Thanks to [Chris K Wensel](https://github.com/cwensel) for the `KATA_HTTP_TIMEOUT` environment setting, project reset-counter support, and early opt-in remote-client mode. - Thanks to [Hugh Brown](https://github.com/hughdbrown) for the per-actor activity digest over kata's event stream. --- ## Project History ### 2026-W25 (Jun 15 - Jun 21, 2026) - Added versioned-release planning, changelog generation, annotated tag creation, tag verification, GitHub artifact workflow, self-update command design, and release documentation. - Added cross-project links that survive issue moves. - Added global daemon selection and fixed named-daemon authentication precedence across CLI, health checks, and TUI paths. - Fixed federation schema-skew quarantine recovery and improved agent recovery when follow-up comments fail after mutations. - Relaxed close throttling so bursty but legitimate sibling closes remain possible with proper evidence. ### 2026-W24 (Jun 8 - Jun 14, 2026) - Added reversible federation leave/rejoin flows across CLI, daemon, and TUI. - Added issue project move support and generated OpenAPI clients. - Documented API compatibility and added `/health` API schema version reporting. - Added generated API artifact drift checks before push. - Fixed optional generated-client response objects and SQLite transient write contention handling. - Improved `kata init --with-agents` handling for `CLAUDE.md`. ### 2026-W23 (Jun 1 - Jun 7, 2026) - Added private-network remote-client mode, bearer-token safeguards, trusted proxy actor headers, simple token identity, and hosted daemon support. - Added federation enrollment UX with actor-bound hub support. - Added `kata init --with-agents`, flat TUI issue lists, OpenAPI schema generation, API docs, and docs screenshot hydration. - Added docs infrastructure, SaaS tracker comparison material, Vercel deployment helpers, and curated design documentation. - Added Windows daemon support, Unix runtime URL publication, daemon lifecycle adoption, telemetry, and storage abstraction groundwork. ### 2026-W22 (May 25 - May 31, 2026) - Built the first hub-and-spoke federation workflows for enrollment, pull/push sync, quarantine, trust-boundary documentation, and local-first project adoption. - Added shared daemon catalogs with TUI switching and trusted private-network bearer auth. - Added cross-project ready queues, ready filters, claim workflows, and stable agent output formatting. - Added project archive/restore behavior and safer list/import handling for larger project data. - Hardened JSONL export/import, backup guidance, and WAL checkpoint documentation. - Added hosted-mode support for `$PORT` platforms and consolidated agent guidance under `AGENTS.md`. ### 2026-W21 (May 18 - May 24, 2026) - Switched the Go module path to `go.kenn.io/kata`. - Continued TUI hierarchy, split-detail, and navigation polish. - Prepared project restore, remote, hosted, and federation work that landed in the following week. ### 2026-W20 (May 11 - May 17, 2026) - Added short issue refs derived from ULIDs in place of per-project numbers. - Added close justification safeguards, anti-abuse guardrails, `--comment` on mutation commands, `kata version`, install docs, and backup/restore docs. - Added daemon metadata, recurrence, auth, and issue move foundations. - Fixed remote-client resolution and schema cutover behavior for pre-existing foreign-key orphans. ### 2026-W19 (May 4 - May 10, 2026) - Added project views, cross-project issue APIs, project archival, early remote-client mode, and the first public PR contributions. - Added TUI split-detail navigation fixes, child graph ordering, list header redesign, opt-in mouse support, marker-gutter polish, and unified detail scrolling. - Added the Beads importer, first-class priority support, and schema cutover model. - Added SQLite lock retries, daemon stale-socket cleanup, module version revision formatting, and percent-encoded git remote init fixes. - Consolidated relationship editing into create/edit flags. ### 2026-W18 (Apr 27 - May 3, 2026) - Bootstrapped kata's local-first architecture: project binding, SQLite storage, daemon API, runtime discovery, Cobra CLI root, and test tooling. - Added issue/comment lifecycle storage, events, project initialization, health checks, lifecycle smoke tests, and daemon handler coverage. - Added relationships, labels, ownership, ready queries, search, idempotency, soft delete, restore, purge, and purge auditing. - Added polling and SSE events, `kata events --tail`, reset handling, and hook execution with queues, worker pools, captured output, rotation, pruning, and reload. - Built the first full TUI, then expanded it with filters, search, editor integration, help, hierarchy rendering, responsive layouts, label chips, split-pane mode, and document-style detail pages. - Added JSONL export/import, schema cutover support, stable UID references, and the federation foundation.