@pellux/goodvibes-agent 1.6.4 → 1.8.0
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 +18 -0
- package/README.md +1 -1
- package/dist/package/main.js +1663 -818
- package/docs/README.md +1 -1
- package/package.json +2 -2
- package/release/live-verification/live-verification.json +12 -12
- package/release/live-verification/live-verification.md +13 -13
- package/src/runtime/services.ts +5 -3
- package/src/runtime/ui/model-picker/types.ts +3 -1
- package/src/version.ts +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -2,6 +2,24 @@
|
|
|
2
2
|
|
|
3
3
|
Product-facing release notes for GoodVibes Agent.
|
|
4
4
|
|
|
5
|
+
## 1.8.0 - 2026-07-09
|
|
6
|
+
|
|
7
|
+
- Adopts platform SDK 1.6.1.
|
|
8
|
+
- When a provider rejects a request as exceeding the model's context window, the agent now compacts immediately and retries once — and learns that endpoint's real context ceiling so future window math, compaction thresholds, and meters stop trusting over-stated catalog values.
|
|
9
|
+
- Agent-completion notices are delivered to the conversation exactly once; they no longer repeat with escalating urgency tags turns after the agent finished.
|
|
10
|
+
- Finished agent/swarm process subtrees can be archived out of the live fleet view into a session archive (and restored), on this machine and over the wire for remote surfaces.
|
|
11
|
+
- Release live verification: a doctor exit whose only findings are operator-configuration risk advisories no longer fails the strict gate; real errors still do.
|
|
12
|
+
|
|
13
|
+
## 1.7.0 - 2026-07-08
|
|
14
|
+
|
|
15
|
+
- Adopts platform SDK 1.5.0.
|
|
16
|
+
- When the model itself reports that its context window filled up while
|
|
17
|
+
- producing a response, the agent now compacts its conversation immediately —
|
|
18
|
+
- before the next model call — instead of waiting for local token estimates
|
|
19
|
+
- to cross a threshold. The model's own report is treated as authoritative.
|
|
20
|
+
- Custom per-model context windows configured on this machine now apply to
|
|
21
|
+
- agent runs too, and persist across restarts.
|
|
22
|
+
|
|
5
23
|
## 1.6.4 - 2026-07-07
|
|
6
24
|
|
|
7
25
|
- The exec tool no longer denies destructive- or escalation-class commands (kill, docker, sudo) that the permission configuration allows — class-level risk is decided by permission settings alone. The only remaining unconditional block is a small frozen catastrophic list (root filesystem deletion, raw disk destruction, fork bombs).
|
package/README.md
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
# GoodVibes Agent
|
|
2
2
|
|
|
3
3
|
[](https://opensource.org/licenses/MIT)
|
|
4
|
-
[](#install)
|
|
5
5
|
|
|
6
6
|
GoodVibes Agent is the installable autonomous operator assistant for GoodVibes. It keeps the existing terminal renderer and workspace bones, but the product goal is different from a vibecoding harness: the user should experience one assistant that can chat, plan, remember, research, schedule, send, generate, run visible agents, and operate the GoodVibes daemon contract with clear confirmation gates.
|
|
7
7
|
|