ur-agent 1.78.7 → 1.78.9

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 CHANGED
@@ -1,7 +1,85 @@
1
1
  # Changelog
2
2
 
3
+ ## 1.78.9
4
+
5
+ - Provider-reported context overflow now triggers one bounded automatic
6
+ compaction-and-retry in the shipped build instead of immediately showing
7
+ `Context limit reached`. Emergency summary input replaces image/document
8
+ payloads with markers so a large Computer screenshot cannot prevent the
9
+ recovery itself; the original transcript remains unchanged. Automatic
10
+ recovery still respects the user's disable setting and never recurses inside
11
+ compact/session-memory workers.
12
+ - Live answer streaming no longer paints partial model self-talk and removes it
13
+ a moment later. The undecided trailing paragraph stays buffered; once
14
+ internal planning is identified, one stable `Reasoning in progress` rail is
15
+ shown until the finalized answer replaces it. Provider-native thinking is
16
+ also private on the normal screen and remains available only in an explicitly
17
+ opened transcript or verbose diagnostics. This is presentation-only and uses
18
+ a monotonic incremental scan, so model context, reasoning depth, and streaming
19
+ throughput remain intact.
20
+
21
+ - Every real user work turn now appears in the live task board immediately,
22
+ even before a model emits `TaskCreate`. The first explicit model task
23
+ atomically replaces that automatic seed, so parallel task creation does not
24
+ duplicate or lose work. A simple successful turn completes its seed; an
25
+ aborted turn leaves it pending for honest recovery.
26
+ - User interruptions no longer archive pending or in-progress tasks into an
27
+ apparently empty list. The next prompt advances the generation while keeping
28
+ unfinished work visible, adds the new request, and instructs the agent to
29
+ reconcile, skip superseded work, and continue updating statuses. Completed
30
+ or otherwise terminal lists still move to readable history.
31
+ - Crew concurrency tests now declare their branches read-only, matching the
32
+ runtime policy that unknown shared-checkout work serializes. The artifact
33
+ HTTP integration test also distinguishes a managed sandbox's blanket socket
34
+ denial from a real occupied-port failure without changing production server
35
+ behavior.
36
+
37
+ - Release tags can no longer be created safely from an old commit while a
38
+ newer version exists only in the working tree. `bun run release:tag` checks
39
+ that the tree is clean, the package version and newest changelog entry are in
40
+ `HEAD`, that exact commit is already the remote branch tip, and the version's
41
+ tag is unused. `--push` then creates and pushes one annotated immutable tag;
42
+ GitHub Actions owns GitHub Release and npm publication from the verified
43
+ tarball. Failed stale tags are left as history and the next patch version is
44
+ used instead of moving a tag.
45
+
46
+ ## 1.78.8
47
+
48
+ - Long model self-deliberation emitted through the ordinary answer channel is
49
+ compacted into a slim `Reasoning condensed` rail. The useful summary and
50
+ result remain visible, `ctrl+o` expands the complete trace, and the original
51
+ text still stays in session history and model context; this is a cached,
52
+ linear-time display change, not reasoning truncation.
53
+ - The `◭ Mashoofing…` activity row remains present for the entire live turn.
54
+ Its parenthesized detail moves through the real phase (`thinking`,
55
+ `requesting`, `responding`, `preparing tool`, or `working`) and continues to
56
+ show elapsed thinking time and token activity instead of disappearing
57
+ between stream states.
58
+ - Large prompts are decomposed into a bounded dependency graph before work is
59
+ delegated. Natural lowercase follow-ups and semicolon clauses split
60
+ correctly, natural phrases such as `agents/subagents` are no longer mistaken
61
+ for paths, and invalid model-produced graphs (missing targets, self-edges,
62
+ cycles, duplicates, or excessive fan-out) fall back to the deterministic
63
+ planner instead of reaching the task store.
64
+ - Parallelism now follows the work, not merely the requested worker count.
65
+ Independent read-only branches may overlap; unknown or shared-checkout
66
+ mutations serialize across tasks, crews, and concurrent top-level plans.
67
+ Mutating subagents may overlap only in explicit isolation, whose worktrees
68
+ start from the exact clean current revision rather than a possibly stale
69
+ remote branch.
70
+ - Task outcome learning is atomic across processes and records only work that
71
+ actually executed. Batched prompt-plan and crew outcomes cannot overwrite one
72
+ another, malformed stores normalize safely, and model routing waits for five
73
+ observations and compares confidence-adjusted success rather than trusting a
74
+ tiny perfect sample.
75
+
3
76
  ## 1.78.7
4
77
 
78
+ - A short approval turn such as `ok`, `proceed`, or `go ahead` keeps the active
79
+ task generation. It previously looked like a brand-new request, so the plan
80
+ was archived just before the agent tried to complete its first task; the next
81
+ dependent creation then failed too because its blocker had disappeared.
82
+ Whole-message matching keeps real follow-up requests on the fresh-list path.
5
83
  - `AskUserQuestion` recovers a flattened option list even when the model omits
6
84
  the outer question text. Eight `{description, header, label}` choices now
7
85
  render under a neutral prompt with every label and description preserved,
package/README.md CHANGED
@@ -785,6 +785,8 @@ bun run release:check
785
785
  bun run package:check
786
786
  npm pack --dry-run
787
787
  npm publish --dry-run
788
+ bun run release:tag # read-only committed/pushed preflight
789
+ bun run release:tag -- --push # immutable tag; CI publishes the artifact
788
790
  ```
789
791
 
790
792
  `dist/cli.js` is intentionally tracked because GitHub installs use the bundled
package/RELEASE.md CHANGED
@@ -86,14 +86,34 @@ Do not publish if the package version already exists on npm.
86
86
 
87
87
  ## Publish
88
88
 
89
- Only after every check passes and the working tree is clean:
89
+ GitHub Actions publishes the exact tarball that passes the release workflow.
90
+ Do not run `npm publish` separately and do not create a tag before the release
91
+ commit is on the remote branch.
92
+
93
+ Only after every check passes, commit the complete release and push it:
90
94
 
91
95
  ```bash
92
96
  git add .
93
97
  git commit -m "chore: polish UR-Nexus production release"
94
98
  git push origin master
95
- npm publish
96
99
  ```
97
100
 
101
+ Wait for the `Test Production` workflow on that commit to pass. Then use the
102
+ guarded tag command; its default mode is a read-only preflight:
103
+
104
+ ```bash
105
+ bun run release:tag
106
+ bun run release:tag -- --push
107
+ ```
108
+
109
+ The command refuses to tag a dirty tree, an uncommitted version, a commit that
110
+ is not yet the remote branch tip, a mismatched changelog, or a local/remote tag
111
+ that already exists. `--push` creates one annotated immutable tag pointing at
112
+ the verified commit. That tag starts `.github/workflows/release.yml`, which
113
+ rebuilds and verifies the package, publishes the GitHub Release, and publishes
114
+ the same verified tarball to npm.
115
+
116
+ If a tag was accidentally pushed from the wrong commit, keep it as failed
117
+ release history and bump to a fresh patch version. Do not move a published tag.
98
118
  If the remote default branch is not `master`, push the checked-out release
99
- branch instead.
119
+ branch instead; the tag command verifies that branch by name.