ur-agent 1.83.2 → 1.84.0

package/CHANGELOG.md

@@ -1,5 +1,47 @@
 # Changelog
 
+## 1.84.0
+
+- Added production npm marketplace sources with scoped-package and version
+  selectors, custom registries, isolated installation, atomic cache promotion
+  and rollback, actionable failures, and complete installed-plugin cache
+  invalidation. Public and technical plugin documentation now describes the
+  same executable contract.
+- Made input-token accounting provider-aware. OpenAI Responses, Anthropic,
+  Gemini, llama.cpp, and vLLM use native non-generating count paths where
+  available; Ollama, OpenRouter, LM Studio, Unsloth, and compatible endpoints
+  use request-shaped local estimation without reusing another provider's
+  fixture or carrying unreachable Bedrock/Vertex branches.
+- Corrected reasoning controls across dynamic providers. Ollama capability
+  discovery now treats ordinary thinking models as native boolean `think`
+  models and sends strings only for explicitly advertised graded ladders or
+  aliases. Added `/thinking on|off|toggle|status`; boolean-capable model pickers
+  use Left for off and Right for on only when the runtime has a real two-state
+  transport. `/effort` never claims a fabricated level, and Ultra remains
+  limited to an explicit provider-advertised beyond-high value. OpenRouter
+  provider selection now reuses its endpoint-scoped live catalogue instead of
+  forcing a slow refresh each time.
+- Replaced incomplete runtime scaffolding with working implementations: a
+  validated tmux-backed Tungsten tool and live monitor; a buildable KAIROS
+  assistant/proactive runtime with transcript persistence, dream lifecycle,
+  scheduling, and communication tools; complete tool output/permission
+  contracts; and a non-React MCP desired-state controller that handles
+  reconnect, authentication, partial failure, and late-connection races.
+- Completed the keybinding and SDK migrations. Shortcut labels now come from
+  the typed registry with correct last-write and Alt/Meta behavior, feature
+  gates hide unavailable bindings, remote tools receive a full fallback
+  contract, and SDK initialization emits canonical `Agent` names while
+  retaining compatibility only at the input boundary.
+- Fixed production-shell and test-isolation defects: `getCwd` is always bound,
+  spawned zsh sessions pass unmatched globs through instead of aborting,
+  missing grep/rg targets become precise model-facing recovery guidance, and
+  secondary-model tests no longer inherit cached live provider settings.
+- Synchronized the slash-command reference from the shipped registry, separated
+  commands from bundled skills and feature-gated internals, expanded KAIROS and
+  provider technical coverage, ignored only runtime-owned `.ur/` state, and
+  added pre-build version consistency enforcement across npm, documentation,
+  IDE extensions, source fallbacks, and the generated CLI.
+
 ## 1.83.2
 
 - Synchronized the public guides, static documentation site, examples, and

package/README.md

@@ -376,9 +376,14 @@ In the interactive app, `/model` is a two-step, provider-first picker:
    their curated model list because the official CLIs expose no models API. The
    generic `subscription` entry is an internal placeholder hidden from listings.
 
-   In the model catalog, use **Up/Down** to browse. The effort row updates to
+   In the model catalog, use **Up/Down** to browse. For graded models, the effort row updates to
    the focused model's capability-backed selectors; use **Left/Right** to cycle
-   only values UR can map to provider-native levels before pressing Enter. `ultra` is UR's visible
+   only values UR can map to provider-native levels before pressing Enter. For
+   boolean-thinking models on runtimes with a native two-state mapping (currently
+   Ollama and direct Anthropic), Left selects off, Right selects on, and `t`
+   toggles. The same state is available directly through `/thinking on|off`.
+   Generic OpenAI-compatible runtimes never receive an invented boolean field.
+   `ultra` is UR's visible
    beyond-high ceiling selector. It appears only when the provider/model
    advertises `ultra`, `max`, `xhigh`, or an explicit equivalent, and the row
    shows the exact mapping (for example, `ultra→max`). Models whose graded
@@ -386,9 +391,13 @@ In the interactive app, `/model` is a two-step, provider-first picker:
    `max` request resolves visibly to that model's highest supported non-Ultra tier
    (commonly `max`, `xhigh`, or `high`), and that resolved value is sent to the provider. llama.cpp models
    are checked lazily through their model-scoped `/props` capability while the
-   cursor moves. Ollama models are checked through `/api/show`; Kimi K3 exposes
-   `low`, `high`, and `max`, and the chosen level is sent through Ollama's
-   native `think` field; Ultra maps to Kimi's advertised `max`. OpenRouter additionally
+   cursor moves. Ollama models are checked through `/api/show`; its `thinking`
+   capability enables boolean thinking without inventing a graded ladder.
+   If `/effort max` or another graded request is used for a boolean-only model,
+   UR enables thinking but reports that no graded level was sent.
+   GPT-OSS uses its documented `low|medium|high` values, while any other graded
+   values or Ultra aliases must be explicitly present in provider metadata.
+   The resolved value is sent through Ollama's native `think` field. OpenRouter additionally
    shows compact model names, FREE/PAID tier, context size, tool/reasoning
    support, and the full untruncated ID immediately below the focused entry.
    Its endpoint-scoped catalog is reused for five minutes; Ctrl+R forces an
@@ -615,6 +624,7 @@ validator, and Markdown language-adapter metadata.
 
 ```sh
 ur plugin list
+ur plugin marketplace add npm:@acme/ur-marketplace@latest
 ur plugin search git
 ur plugin search --capability skills --json
 ur plugin show github@ur-plugins-official
@@ -636,6 +646,13 @@ installed, project, and bundled plugin manifest and reports its declared
 components and capability surface, so you can review what a plugin touches
 before enabling it.
 
+Marketplace catalogs can come from GitHub/Git, direct URLs, npm packages,
+local files/directories, or inline settings. npm sources use
+`npm:<package>[@<version-range-or-dist-tag>]`, respect the user's npm registry
+and authentication configuration, and must ship `.ur-plugin/marketplace.json`.
+See the [Plugin Guide](docs/plugins.md) for private-registry configuration and
+refresh behavior.
+
 The npm package includes `README.md`, `QUALITY.md`, `docs/`, `documentation/`,
 and `plugins/`, so the npm package page and installed artifact both carry the
 marketplace documentation, core plugins, community staging directory, and