ur-agent 1.60.1 → 1.61.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +35 -0
- package/dist/cli.js +1133 -850
- package/docs/USAGE.md +4 -0
- package/documentation/index.html +1 -1
- package/extensions/jetbrains-ur/build.gradle.kts +1 -1
- package/extensions/vscode-ur-inline-diffs/package.json +1 -1
- package/package.json +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,40 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## 1.61.1
|
|
4
|
+
|
|
5
|
+
- Fixed secondary model queries failing on any Ollama setup whose session model
|
|
6
|
+
is not `qwen2.5-coder:7b`. `getSmallFastModel()` fell back to the compiled
|
|
7
|
+
default when auto-routing was off or no model list had been discovered, so a
|
|
8
|
+
user running `kimi-k2.7-code:cloud` had every WebFetch summarisation and
|
|
9
|
+
classifier call rejected with "Model qwen2.5-coder:7b is not available for
|
|
10
|
+
provider ollama". A fallback to a model the user may never have pulled is a
|
|
11
|
+
guaranteed failure; it now falls back to the session model, which is
|
|
12
|
+
guaranteed to exist.
|
|
13
|
+
- WebFetch no longer returns a failed summarisation as if it were the page.
|
|
14
|
+
The error text was passed through as the fetch result, so the evidence ledger
|
|
15
|
+
recorded "API Error: Provider ollama ..." as the content of example.com and a
|
|
16
|
+
`--check` against real page text correctly found nothing. It now throws, and
|
|
17
|
+
the failure is visible as a failed tool call.
|
|
18
|
+
|
|
19
|
+
## 1.61.0
|
|
20
|
+
|
|
21
|
+
- Added `ur selftest`, end-to-end drills for the gap that produced every
|
|
22
|
+
serious defect in recent releases: a module that is correct while something
|
|
23
|
+
between it and the user is not — a wire format, a CLI registration, an exit
|
|
24
|
+
code, a wording choice. The drills deliberately do not import the modules;
|
|
25
|
+
they spawn the shipped binary against real directories and assert on what a
|
|
26
|
+
user would see.
|
|
27
|
+
- Five drills run automatically and cover the four defects that reached you:
|
|
28
|
+
memory integrity reporting an unchecked store as verified, the trajectory
|
|
29
|
+
gate printing FAILED while exiting 0, `--costs` hiding which directory it
|
|
30
|
+
searched, and per-agent attribution itself.
|
|
31
|
+
- Drills that need a live model cannot be automated, so they are emitted as
|
|
32
|
+
prompts with the specific observation that separates working from broken —
|
|
33
|
+
"it seemed fine" is how the screenshot bug survived its first report.
|
|
34
|
+
- `ur selftest run` exits non-zero on failure, and a test asserts the drills
|
|
35
|
+
actually fail when pointed at a broken binary. A self-test that cannot fail
|
|
36
|
+
is decoration.
|
|
37
|
+
|
|
3
38
|
## 1.60.1
|
|
4
39
|
|
|
5
40
|
- `ur memory-integrity verify` no longer reports an empty or missing store as
|