v0.1.0

0.1.0 (2026-10-20)

⚠ BREAKING CHANGES

• config: DefaultConfig has no provider or model, and LoadConfig no longer requires them; run Validate on the merged config.
• registry: Totem no longer includes Init/Cleanup; implement Initializer (Init(ctx, eng)) and Cleaner instead, and run totems with registry.Invoke.
• config: Config.Passes and Config.Dimensions are removed; use Config.Totems / TotemConfig.
• engine: Agent.Run takes a RunEnv; NewReAct takes no config and ReActConfig is gone; PermissionConfig must be changed through SetTool / SetDefault once handed to an Engine.

Features

• a purple planchette logo, built on the original design (edee6f4)
• bootstrap: expose config-to-engine wiring for external totems (8853598)
• bootstrap: run a totem binary from one line with bootstrap.Main (993f6e1)
• chat: name the totems on the welcome again (f218297)
• chat: redesign the TUI — welcome, tools, diffs, approvals, search (e15f34c)
• chat: run every totem, /learn, command output and golden screens (bb40b5c)
• chat: smooth streaming, @file mentions, /undo, /task, sessions and themes (4b64c28)
• chat: trim the welcome to the logo and four places to start (ff1be74)
• cli: add a shared palette and restyle doctor, setup and init (e66bcd8)
• cli: draw totem runs live on the terminal (ab58dbb)
• config: require the developer to choose the provider and model (bd20015)
• config: require trust before applying a discovered project config (38482fb)
• engine: add an undo journal and project file listing (d8c2374)
• engine: confine bash commands with Linux Landlock (opt-in) (df96861)
• engine: fit tool results and history to the model's context window (fdea40d)
• engine: report run start/end, turn tool summaries and context use (72d1b10)
• generate: scaffold standalone external totem modules (5f7daac)
• harness: add CLI, MCP server, engine, providers and totems (852ae3d)
• mcp: run tool calls concurrently and support cancellation (5833612)
• models: list each provider's models from its own API (9f6186c)
• registry: make the totem lifecycle optional and add typed inputs (6c1c5b7)
• sticker: mark the sticker with the new planchette (2eeb850)
• take the context window from the config, the model lists or Ollama's default (8e314fc)
• theme: add catppuccin, dracula and mono themes (5245d54)
• theme: add OUIJA_NO_ANIMATION for a still screen (c6f3d55)
• theme: move ouija's palette from pink toward purple (7609d13)
• trace: mark turns instead of resetting the tracer (cc18514)
• trace: show the grep/glob pattern next to the path (c9e5807)

Bug Fixes

• bash: judge expanded words in the denylist and keep failing output (85c205a)
• bash: stop learned patterns from pre-approving extended commands (208c0a5)
• build: pass ldflags once in make install and trim release paths (52f47dc)
• chat: fetch model lists off the event loop, keep text terminal-safe (cf824a8)
• chat: keep the session trace and match approval keys to the menu (65d6ccd)
• chat: send slash commands as typed and fix approval and queue flow (5ff153a)
• chat: send the assembled system prompt and unify generation handling (7c4ad1f)
• chat: show the checkboxes of a task list (f8b385c)
• chat: Windows key releases, editor default and config writes (6f26e15)
• cli: make --help safe, exit codes consistent and first-run hints right (66ee37f)
• cli: reject stray totem args and surface broken config errors (2415db7)
• cli: stop listing the task totem twice in help (1a06630)
• cli: strip terminal controls from repo and model text; check the key (8698e69)
• config: harden project trust and never overwrite a broken config (99fb4e3)
• config: keep the sandbox root inside the project (60859ea)
• config: scope the project API key to its provider (b62fb49)
• deps: build with Go 1.26.9 for 27 standard-library vulnerabilities (fc8ef01)
• doc: lock the generated HTML down with a Content-Security-Policy (01b3023)
• doc: put the CSP before any model-written markup (40a5134)
• engine: bound provider streams by silence, not length, and retry dropped ones (289ff54)
• engine: classify stream errors and harden the tool pipeline (de5d87b)
• engine: close sandbox escapes through dangling symlinks (2ccc5bf)
• engine: close Windows sandbox escapes and run bash in a real shell (19ae53a)
• engine: enforce protected paths through symlinks (7c3fe4d)
• engine: guard protected paths and the home dir on Windows too (69199c1)
• engine: judge bash path words by value and physical location (66ae8e1)
• engine: keep tool order, count cache tokens and stop on cancel (9e09a99)
• engine: make structured output and long chats reliable (6dea3e9)
• engine: make the file tools work on large repositories (442c3f6)
• engine: mark a failing bash command failed and keep its output (ab0abbb)
• engine: never copy the resolved API key into a project config (bcbe8b7)
• engine: offer the model only the tools the gate could run (004727a)
• engine: protect ouija's global config and project config files (68969e2)
• engine: refuse cut-off structured answers, flag git hook changes (39d74df)
• engine: require every simple command to match the bash allowlist (b0a90ac)
• engine: stop the bash path scan from panicking and judge globs and $HOME (fce92d0)
• generate: anchor generated .gitignore entries (823473b)
• generate: pin ouija only for real releases in the external scaffold (b15c316)
• generate: reject unbuildable totem names and roll back failures (3586c61)
• learn: keep the project's knowledge at its root (7cb3c08)
• mcp: do not panic installing into a config with "mcp": null (f0e4015)
• mcp: honour an explicit config path in tool calls (ddc83b1)
• mcp: stop on signals, ignore notifications, harden resources/read (daceee7)
• models: scope the list cache and offer only chat models (3ebbc30)
• models: stop advertising OpenCode's free tier (540ea66)
• providers: assemble Anthropic tool input and harden streaming (12174ba)
• providers: cap the server-suggested Retry-After wait (1f90aea)
• providers: emit OpenAI tool calls in order and classify failures (49194fe)
• providers: report refused keys as auth errors with an actionable hint (eb8e07f)
• providers: use native structured outputs on Anthropic and declare tools in the final call (e0b7770)
• registry: make Invoke safe for concurrent hosts and type --set inputs (10870ce)
• resolve totem outputs and knowledge through the sandbox (a567918)
• sandbox: require Landlock ABI 9, confine scopes, guard grants (aa1f3bb)
• setup: keep trusted projects and unknown keys in the global config (1a79389)
• task: read knowledge and write the plan under the project root (3c922fc)
• totems: escape sticker SVG, never overwrite user docs, bound PDF (6d7bf36)
• totems: never save an empty plan, fill the doc cover, fix sticker names (f6ee8f8)
• write config files atomically (31d63b4)

Documentation

• architecture: show the public packages in the layer tree (1e8df07)
• document git guard, typed --set, totem concurrency and doc cover keys (7ca3703)
• document trust, allowlist and sandbox boundaries (3f3c4a2)
• fix first-run docs (config paths per OS, flags, versions, trust keys) (b20ef2f)
• invite totem contributions and show the ecosystem roadmap (aabe3a0)
• make totems the lead story for developers (531678b)
• new logo and banner, and a demo GIF in the README (57175a0)
• pitch own-agent capability in the README hero (8783ad2)
• readme: refresh the screenshots and the GIF for the new welcome (e4dc2c0)
• readme: refresh the screenshots and the GIF in the purple palette (ac072ab)
• readme: show the chat with screenshots (e694907)
• resolve totem writes with SafeResolve and document enginetest (f170861)
• restructure into Diátaxis and rewrite root guides (355f16f)
• restructure README for a developer's first visit (bc3a44d)
• show chat as daily driver in quick start (e5b02dc)

Refactoring

• config: move totem settings into per-totem config sections (e7cf5ab)
• engine: snapshot each run's environment and make the Engine safe for concurrent use (2310a69)