ur-agent 1.44.10 → 1.45.3

package/CHANGELOG.md

@@ -1,5 +1,113 @@
 # Changelog
 
+## 1.45.3
+
+- Made slash-command resolution deterministic across bundled skills, plugins,
+  project skills, workflows, and built-ins. Duplicate canonical tokens are
+  rejected by source priority and conflicting aliases are removed; registry
+  tests now verify every shipped command token, description, and lazy loader.
+- Removed overlapping `/paper`, `/security`, `/audit`, `/skills`, and
+  `/sandbox` registrations. The single `/sandbox` command now provides both
+  interactive settings and `status`, `check`, `init`, `eval`, and `exclude`
+  subcommands.
+- Added `/ci-loop --cwd <path>` and working-directory evidence. Test-runner
+  "No tests found" failures now stop after one attempt with actionable cwd
+  guidance instead of invoking a fix agent repeatedly.
+- Serialized concurrent artifact-viewer startup so simultaneous callers share
+  one server and one reported URL.
+
+## 1.45.2
+
+Correctness and containment release completing the runtime audit.
+
+- Made sandbox, security-scope, WebFetch, API, browser, database, test-runner,
+  GitHub mutation, lifecycle-hook, and file-edit boundaries fail closed. URL
+  validation now covers DNS resolution and every redirect; browser navigation
+  uses a persistent Playwright session with guarded subrequests.
+- Completed provider request mapping and error handling across OpenAI,
+  Anthropic, Gemini, OpenRouter, Ollama, and OpenAI-compatible runtimes. Generic
+  compatible endpoints use their own credential key, offline mode blocks cloud
+  dispatch, retry zero is honored, and local Ollama is the default route.
+- Made AST/LSP workspace edits transactional and containment-safe, including
+  symlink rejection, stale-edit preconditions, atomic writes, new-file rollback,
+  and TypeScript-language-service import organization.
+- Hardened SSE/WebSocket replay, transcript write serialization, direct-connect
+  cancellation, arena isolation/judging, thread sharing, and same-timestamp file
+  replacement detection. Session/diff/lab/security-fix paths reject traversal
+  and symlink escapes.
+- Fixed VS Code multi-root selection, stale turn callbacks, and diff manifest
+  validation. Replaced the JetBrains plugin's nonexistent `/v1/prompt` call
+  with project-scoped JSON-RPC sessions over `/acp`; the plugin builds against
+  IntelliJ IDEA 2024.2.
+- Enabled transcript deep search instead of the compiled-off path. Worktree
+  skills now keep changes local, ask before the final full verification suite,
+  and never commit, push, or open a PR unless explicitly requested. The
+  `autoApprove` permission mode is unchanged.
+- Removed an internal SDK barrel's throw-only runtime exports; the supported
+  programmatic contract remains `ur -p` stream-json. Provider fallback is now
+  documented and reported accurately as explicit recovery guidance, never an
+  automatic cross-provider switch.
+
+## 1.45.1
+
+Completes the three partially-delivered 1.45.0 items to 100%.
+
+- Semantic code search is now zero-config: the CodeSearch tool auto-enables
+  the moment a built index exists (`ur code-index build`) — no `UR_CODE_INDEX`
+  env var needed. The env var remains an override in both directions
+  (`UR_CODE_INDEX=off` disables even with an index present).
+- JetBrains plugin now actually builds: migrated to the IntelliJ Platform
+  Gradle Plugin 2.x (the 1.x plugin is incompatible with Gradle 9) and
+  compiled against IntelliJ IDEA Community 2024.2, producing a distributable
+  zip via `gradle buildPlugin`.
+- Deprecated top-level `disableAutoMode` now emits a one-time warning at
+  load pointing to `permissions.disableAutoMode` (both keys still honored;
+  removal reserved for the next major).
+
+## 1.45.0
+
+Top-tier feature release — closes the gaps against 2026's leading agents.
+
+- `ur cloud` — detached best-of-N tasks (the local-first codex-cloud
+  analogue): `run "<task>" --attempts N` races up to 8 isolated worktree
+  agents in the background via the arena judge; `list`/`show <id>` browse
+  results any time; `apply <id>` applies the winning diff. Outcomes feed the
+  automatic learning store.
+- `ur wiki` — living repo wiki (Devin-Wiki analogue): `generate` writes
+  `.ur/wiki/` (overview, architecture by import in-degree, dependency map)
+  from project DNA + the code index; `install-hook` refreshes it after every
+  merge; `map` maintains `.ur/repo-map.md`.
+- Repo map in context (Aider/Cursor pattern): when `.ur/repo-map.md` exists
+  and is fresh (<7 days), a byte-capped orientation map is injected into the
+  system prompt. Zero tokens until you generate one.
+- `ur recipe` — structured-output playbooks (Devin-style): stored prompt +
+  JSON Schema; `run` spawns a child session whose final answer must validate
+  (one automatic repair round with the validation errors).
+- `ur thread share` — share a session transcript as a local web page,
+  served at `/threads/<id>` on the artifacts server. Local-first: nothing
+  leaves the machine unless you expose the port.
+- Web dashboard at `/dashboard` on the artifacts server: cloud tasks,
+  background agents, live task board, and learning stats on one page
+  (`/api/dashboard` for JSON).
+- `ur audit export` — hash-chained audit trail (tool actions + run traces)
+  as JSONL/CSV; `ur audit verify <file>` proves an export wasn't edited or
+  reordered.
+- `/pdf <file> [pages]` — deps-aware PDF ingestion (pdftotext/pdfinfo) with
+  page ranges, mirroring /image.
+- Crew dynamic fan-out: `ur crew run <name> --dynamic [--max-workers N]`
+  scales the worker pool to the task board (tasks appended mid-run are
+  picked up), governed by a hard concurrency cap.
+- Auto-skillify: `/learn stats` now surfaces skill candidates — categories
+  with ≥5 recorded successes and no matching skill get a concrete
+  /create-skill suggestion.
+- Voice mode and the computer-use MCP server (`ur --computer-use-mcp`) now
+  ship in release builds (previously compile-time-disabled). Voice's native
+  audio backend stays optional and degrades gracefully when not installed.
+- JetBrains plugin scaffold under `extensions/jetbrains-ur/` (experimental,
+  thin ACP client mirroring the VS Code extension).
+- Settings: top-level `disableAutoMode` is marked deprecated in favor of
+  `permissions.disableAutoMode` (both still honored).
+
 ## 1.44.10
 
 - Render `AskUserQuestion` permission requests inside a `PermissionDialog` so

package/README.md

@@ -214,7 +214,7 @@ Development build detected. To update, pull latest source or install from npm.
 Choose a model or Ollama host:
 
 ```sh
-ur --model qwen3-coder:480b-cloud
+ur --model qwen2.5-coder:7b
 ur --ollama-host http://192.168.1.50:11434
 ur --discover-ollama
 ```
@@ -252,12 +252,16 @@ ur config set provider openai-api
 ur config set provider anthropic-api
 ur config set provider gemini-api
 ur config set provider openrouter
-ur config set model qwen3-coder:480b-cloud
-ur provider select-model ollama qwen3-coder:480b-cloud --json
+ur config set model qwen2.5-coder:7b
+ur provider select-model ollama qwen2.5-coder:7b --json
 ur config set base_url http://localhost:11434
 ur config set provider.fallback ollama
 ```
 
+`provider.fallback` records a recovery provider for `ur provider doctor`
+guidance. UR never switches providers automatically: inspect the failure and
+select the recovery provider explicitly with `ur config set provider <id>`.
+
 Provider config accepts canonical IDs and common aliases. Examples:
 `openai-api`, `anthropic-api`, `gemini-api`, `openrouter`, `ollama`,
 `lmstudio`, `LM Studio`, `llama.cpp`, `vllm`, and the subscription CLIs
@@ -358,7 +362,7 @@ as first-class subcommands in the shipped CLI.
 | `ur spec` | Default spec-first workflow: create requirements, design, and task documents under `.ur/specs/`, run the task list, and verify with strict proof gates. |
 | `ur escalate` | Plan or run work with fast and oracle model tiers selected from local model capabilities. |
 | `ur arena` | Run multiple agents on the same task in isolated worktrees and surface the winning diff. |
-| `ur ci-loop` | Run a build or test command, hand failures to a fix agent, and retry with a bounded budget. |
+| `ur ci-loop` | Run a build or test command in an explicit working directory, hand failures to a fix agent, and retry with a bounded budget. |
 | `ur test-first` | Detect the project stack, run compile/test/lint commands, store failure traces, and install edit-time verify gates. |
 | `ur safety` | Inspect or initialize project shell safety policy and evaluate command risk before execution. |
 | `ur sandbox` | Inspect and manage the sandbox/permission architecture: status, dependency check, policy init, and command approval levels. |
@@ -469,9 +473,10 @@ ur ide config vscode        # VS Code / Cursor / Windsurf setup
 
 VS Code, Cursor, and Windsurf connect through the UR Inline Diffs extension;
 Zed and ACP-capable Neovim clients connect through the stdio Agent Client
-Protocol. **JetBrains is not implemented in this repository** — no plugin
-ships from here; only detection code for a future one exists. Start an ACP
-surface with:
+Protocol. The experimental bundled JetBrains plugin under
+`extensions/jetbrains-ur` uses project-scoped JSON-RPC sessions over the
+loopback HTTP ACP endpoint and is installable from its Gradle `buildPlugin`
+zip. Start an ACP surface with:
 
 ```sh
 ur acp stdio                        # stdio ACP agent for editors (Zed, Neovim)
@@ -482,7 +487,9 @@ ur acp status
 See the [IDE Guide](docs/IDE.md) and [ACP Guide](docs/ACP.md) for per-editor
 setup, supported features, and limitations.
 
-New slash skills run agentic work in isolated git worktrees with clean commits and PR output:
+New slash skills run agentic work in isolated git worktrees and leave changes
+local for review. They ask before the final full verification suite and do not
+commit, push, or create PRs unless explicitly requested:
 `/debug-v2`, `/refactor`, `/paper-implementation`, `/benchmark`, `/security-review`, `/dockerize`, `/latex-paper`.
 Install matching agent templates with `ur agent-templates install`.
 
@@ -549,7 +556,7 @@ ur context-pack remember --accepted "Use p-map for bounded concurrency" --scope
 ur context-pack remember --rejected "Switch to esbuild" --alternative-to "Keep bun bundle"
 ur context-pack remember --attempt "Tried Deno runtime" --status superseded
 ur context-pack compress
-ur ci-loop --command "bun test" --max-attempts 3 --dry-run
+ur ci-loop --command "bun test" --cwd . --max-attempts 3 --dry-run
 ur bg run "fix the flaky parser test" --worktree --dry-run
 ur automation create nightly --schedule "0 9 * * 1-5" --prompt "Review open tasks"
 ur repo-edit preview rename oldName --to newName