BAKED 0.3.0

The desktop installers (baked-desktop-*) are not code-signed yet: Windows SmartScreen and macOS Gatekeeper warn on first launch. See docs/getting-started.md.

What's Changed

Dependencies

• Bump actions/checkout from 6.0.2 to 7.0.1 in the actions group across 1 directory by @dependabot[bot] in #54
• Bump the minor-and-patch group across 1 directory with 14 updates by @dependabot[bot] in #72
• Bump the minor-and-patch group with 11 updates by @dependabot[bot] in #281

Changes

• Tell test tasks to report defects outside their scope, not block on them by @Gulliby in #163
• Split a failed run-level check into findings per file owner by @Gulliby in #164
• Send plan-only analysis findings back to the planner once by @Gulliby in #169
• Repair only high and critical review findings after the first round by @Gulliby in #168
• Rename the main agent from Forge to Bake by @Gulliby in #171
• Let analyze replan a coverage gap that names the criteria it misses by @Gulliby in #173
• Repair a medium convergence gap that keeps a criterion failing by @Gulliby in #175
• Desktop: own title bar, New chat in the sidebar, no icon rail by @Gulliby in #177
• Desktop: Tailwind, shadcn components and neobrutalist tokens by @Gulliby in #180
• Desktop: neobrutalist title bar and sidebar by @Gulliby in #181
• AGENTS.md: ship major features on every app surface by @Gulliby in #187
• Drop the ast_edit hint from edit tools offered to workflow agents by @Gulliby in #182
• Order test tasks after the code tasks they cover by @Gulliby in #186
• Settle blocking questions with recorded defaults when the policy assumes by @Gulliby in #188
• Desktop: neobrutalist project home by @Gulliby in #189
• Run typecheck with bun check and drop the shadow CI step by @Gulliby in #192
• TUI: undo a run's uncommitted task edits from the run screen by @Gulliby in #196
• Continue a task attempt that stops at its cap after editing by @Gulliby in #195
• Desktop: neobrutalist chat by @Gulliby in #193
• Stop a workflow when the model provider refuses the account by @Gulliby in #198
• Desktop: neobrutalist runs, Needs you, setup, usage and projects by @Gulliby in #194
• Desktop: undo a run's uncommitted task edits from the run screen by @Gulliby in #200
• Run start's model preflight through the doctor's model checks by @Gulliby in #199
• Keep the run's feature ID when a spec names another one by @Gulliby in #207
• Split oversized test tasks by file and raise the budget before a re-plan by @Gulliby in #208
• Fail a workflow over to another provider when the account is refused by @Gulliby in #203
• Resume at specify when an answered finding refers to the spec by @Gulliby in #210
• Copilot: prompt cache markers for Claude models by @Gulliby in #212
• Give every attempt a step floor on its token cap by @Gulliby in #214
• Let a check name a fixer that implementers run with apply_fix by @Gulliby in #216
• Keep a capped batch task's edits for its continuation by @Gulliby in #218
• Restart a batch task once when its worker is lost by @Gulliby in #220
• Desktop: neobrutalist checkboxes and selects, no-project state on setup screens by @Gulliby in #223
• Cache workflow agent prompts on the Claude API by @Gulliby in #229
• Add a Claude Code provider that keeps the prompt cache across steps by @Gulliby in #230
• Keep other BAKED processes' LSP servers alive when reaping orphans by @Gulliby in #234
• Share one TUI listener per keyboard and resize event by @Gulliby in #222
• Choose assume or ask for spec questions per run and show assumptions at approval by @Gulliby in #232
• Codex provider: keep one app-server thread per conversation by @Gulliby in #213
• Keep finished tasks when a revision only adds acceptance criteria by @Gulliby in #237
• Send blocking analysis findings back to specify or plan in v2 runs by @Gulliby in #238
• Preload task files and trim the implementer prompt by @Gulliby in #242
• Switch a stopped run to another model without losing finished work by @Gulliby in #239
• Keep review findings on unnamed standard edge cases as notes by @Gulliby in #244
• Share one code-intelligence worker between chat and workflow runs by @Gulliby in #245
• Run a failed check's fixer before creating a repair task by @Gulliby in #250
• One-file run report, used as the delivery PR body by @Gulliby in #248
• Notify when a workflow run waits for a person by @Gulliby in #247
• Switch an initiative's remaining runs to another model in one command by @Gulliby in #253
• TUI routing screen for modelRouting (R3) by @Gulliby in #256
• Start a workflow run from a GitHub issue by @Gulliby in #252
• @ mentions and !commands in the chat input by @Gulliby in #258
• Codex provider: native tool calls and streaming over app-server by @Gulliby in #257
• Desktop: sign in to Claude Code and Codex from Settings by @Gulliby in #260
• Give the planner the auto approval limits by @Gulliby in #263
• Desktop: edit every model route in Settings, through a validating routing endpoint by @Gulliby in #261
• Size plans by docs kind, write action and project history by @Gulliby in #265
• Fork a session from a message with file rollback by @Gulliby in #255
• Require a reproducible example before a review finding starts a correction by @Gulliby in #267
• Desktop chat: Compact now and per-session diff stats by @Gulliby in #275
• Compose input in $EDITOR, /undo and /redo, CI check for conflict markers by @Gulliby in #270
• Allow MCP tools in workflow profiles through tool-policy grants by @Gulliby in #278
• Add a Fix CI workflow preset by @Gulliby in #279
• Show per-file diff line stats in the TUI and headless CLI by @Gulliby in #282
• Fix blank New run screen in the desktop mock preview by @Gulliby in #283
• Chat undo/redo in the local API, desktop app and baked session by @Gulliby in #285
• Plan mode and plan review in desktop chat and the local API by @Gulliby in #286
• Project memory from earlier runs for research, spec and planning by @Gulliby in #287
• Preload the spans of symbols a task names in large written files by @Gulliby in #292
• View and edit run memory in /memory, the desktop app and the CLI by @Gulliby in #295
• Desktop: build every screen from components/ui and drop the per-screen CSS by @Gulliby in #276
• Desktop chat: image attachments by @Gulliby in #294
• Oven logo and BAKED icon set in the desktop app by @Gulliby in #303
• Desktop: workflow runs as a conveyor, New run in three steps by @Gulliby in #305
• Add the bench loop orchestrator spec by @Gulliby in #307
• Code graph: desktop section, terminal /graph view and /v1/graph API by @Gulliby in #308
• TUI: neobrutalist bun mascot, boot splash, busy bar and motion setting by @Gulliby in #301
• Desktop: brand scrollbars, single empty Needs you, sidebar order by @Gulliby in #336
• Check spec coverage in the specify validate hook by @Gulliby in #337
• Keep the analysis verdict when the plan-size question is answered by @Gulliby in #338
• Desktop Settings: one page per section, edit in a dialog, plain-language hints by @Gulliby in #339
• Desktop: browse, install and remove project and global skills by @Gulliby in #340
• Give research a step floor, finalize earlier and retry an unsubmitted attempt by @Gulliby in #343
• Auto-approve greenfield manifests and do not count host-split tasks by @Gulliby in #344
• Host-filled, validated submission path for review outputs by @Gulliby in #345
• Add one review bundle diff to the review snapshot by @Gulliby in #349
• Add a request-coverage gate for described programs on empty workspaces by @Gulliby in #348
• Pick workflow routing models from the list of available models by @Gulliby in #351
• Update bench loop specs after round 2 by @Gulliby in #352
• Publish verified bench apps automatically by @Gulliby in #353
• Add the Claude settings template to the bench loop spec by @Gulliby in #354
• Add updates to existing projects to the bench loop by @Gulliby in #355
• Make BAKED-side improvement the bench loop objective by @Gulliby in #356
• Assign finding ids on the host by @Gulliby in #366
• Build host Workspace Facts instead of the greenfield label by @Gulliby in #367
• Defer manual criteria in required reviews by @Gulliby in #368
• Fix the analysis and clarification loop by @Gulliby in #369
• Approve manifest edits on a scaffold baseline by @Gulliby in #370
• Workflows on by default; set up a project when a folder is opened by @Gulliby in #342
• Continue the run after a headless decision by @Gulliby in #371
• Make task worktrees byte-exact on Windows and classify refused edits by @Gulliby in #392
• Give code tasks the static checks that cover their files and fix before checking by @Gulliby in #393
• Warn when class routing overrides --model; add --model-strict by @Gulliby in #396
• Do not discharge a surviving inconsistency as an assumption by @Gulliby in #410
• Repair once per failing check run and accept a no-op repair whose checks pass by @Gulliby in #412
• Expand custom slash commands in API and desktop chat by @Gulliby in #414
• Re-prompt an inconclusive required reviewer once and route its gaps to converge by @Gulliby in #423
• Desktop: recover an interrupted effect, fork a run, discard a worktree by @Gulliby in #420
• Chat settings schema and GET|PATCH /v1/settings by @Gulliby in #415
• Desktop chat settings pages, TUI pickers on the settings schema by @Gulliby in #424
• Rename a project from the API, the desktop app and the TUI by @Gulliby in #427
• Headless CLI: usage, attention, session list/delete, --remove-key, login/logout by @Gulliby in #426
• Accept --model-strict in the workflow argv parser by @Gulliby in #433
• Long-term chat memory in the local API and the desktop app by @Gulliby in #418
• Desktop: show each stage's model on New run, add the Windows taskbar badge by @Gulliby in #429
• Headless CLI: baked workflow routing set, clear, review and show by @Gulliby in #435
• Switch telemetry and edit run notifications from the TUI and CLI by @Gulliby in #428
• Desktop Git section over a new serve API, shared with the TUI Git menu by @Gulliby in #430
• Headless CLI: baked session export by @Gulliby in #437
• Manage initiatives fully from the TUI and the desktop app by @Gulliby in #416
• Instruction files: switch sources in the desktop, view and edit files in the TUI by @Gulliby in #432
• Feature registry and a parity test across surfaces by @Gulliby in #422
• Give repair tasks a probe and count failing output as progress by @Gulliby in #444
• Desktop: view hooks and switch them off for the session, switch the Codex account by @Gulliby in #436
• Grow the run budget toward its ceiling while tasks finish by @Gulliby in #445
• MCP servers in the local API, desktop app and baked mcp by @Gulliby in #421
• Invalidate only the dependents that read a file the repair changed by @Gulliby in #446
• Delete and export chats in the desktop by @Gulliby in #425
• Ask for a decision when convergence stops instead of blocking by @Gulliby in #447
• Accept a no-op rerun of an invalidated planned task by @Gulliby in #448
• Diagnostics in serve and the desktop: storage cleanup, model events, error log by @Gulliby in #434
• Pass a low partial finding with notes when behavior is verified by @Gulliby in #449
• Run tests on Windows and macOS; enforce acceptance gate ports outside Linux by @Gulliby in #411
• Key convergence repeats on the finding location, pass test-only findings with notes by @Gulliby in #455
• claude-code resilience: restart on a lost tool, OAuth collision backoff, bounded owner_lost resume by @Gulliby in #456
• Reject method manual for spec criteria about repository files by @Gulliby in #457
• Rerun documentation writers and import-only dependents only after a host check by @Gulliby in #458
• Defer a task check that fails on files of tasks still to run by @Gulliby in #459
• Show accepted limitations in the status JSON and run screens by @Gulliby in #460
• Bench loop: Go, .NET, Rust and Java stacks, 3 lanes, difficulty rule by @Gulliby in #469
• Let the request win over a restriction the spec adds under clarification assume by @Gulliby in #470
• Key the findings ledger by repeatOf and let the assume rule count distinct findings by @Gulliby in #472
• Restart a blocked implementer once in a fresh session on host facts by @Gulliby in #471
• Settle superseded unknown effects at acceptance and ask at the stop by @Gulliby in #473
• Rerun a planned task by supersession with host-first checks by @Gulliby in #474
• Abort Claude Code native tool calls and correct the step in the same session by @Gulliby in #483
• Keep each file's line endings in edit_file and multi_edit by @Gulliby in #482
• Run the fixer of a failed task check in the host before a repair task by @Gulliby in #485
• Add Go, .NET, Rust and Maven gates, manifests and a language reviewer by @Gulliby in #486
• Continue an agent that ended on prose and record the implementation from host facts by @Gulliby in #484
• Bench loop: check language servers per round by @Gulliby in #487
• Desktop: recent runs under Runs instead of Acceptance by @Gulliby in #488
• Desktop: cache screen data per project and refresh it in the background by @Gulliby in #491
• Treat a model at capacity as transient with a longer backoff by @Gulliby in #490
• Regression test for racy checkpoint snapshots by @Gulliby in #497
• Derive the requirement ids of planned tasks from their criteria by @Gulliby in #489
• Make the README user-friendly and add desktop screenshots by @Gulliby in #504
• Test the Copilot prompt cache against a fake backend by @Gulliby in #506
• Language server status and settings in the desktop app, the local API and headless CLI by @Gulliby in #495
• Desktop smoke test of the packaged app on Windows and macOS by @Gulliby in #508
• Rework the desktop Quick task form; budget presets applied on the server by @Gulliby in #493
• Bench loop: use headless baked lsp commands for language servers by @Gulliby in #514
• Enforce the feature/ and bugfix/ branch rule for every author by @Gulliby in #511
• Desktop UI tests: renderer in Chromium against the mock bridge by @Gulliby in #507
• Desktop: centre dropdown text on the chevron and use the app font by @Gulliby in #517
• Detect existing GitHub Copilot and GitHub Models logins by @Gulliby in #502
• Reword a chat memory in the TUI /memory browser by @Gulliby in #516
• Desktop UI tests: Electron app against a fake baked serve by @Gulliby in #519
• Scripted model for baked serve and an end-to-end desktop chat with approval by @Gulliby in #520
• Let Bake answer questions about BAKED from bundled docs by @Gulliby in #492
• Release 0.3.0 by @Gulliby in #512

New Contributors

• @dependabot[bot] made their first contribution in #54

Full Changelog: v0.2.0...v0.3.0