@agentskit/harness 0.11.0 → 0.12.0
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 +20 -0
- package/capabilities/public-surface.json +86 -84
- package/dist/cli.js +54 -9
- package/dist/cli.js.map +1 -1
- package/dist/index.d.ts +20 -1
- package/dist/index.js +49 -10
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
- package/release/manifest.json +3 -3
- package/release/notes.md +13 -0
package/CHANGELOG.md
CHANGED
|
@@ -1,5 +1,25 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## [0.12.0] - 2026-09-13
|
|
4
|
+
|
|
5
|
+
Closes gaps found reusing Orca instead of reinventing it. Orchestration mutations (`run-create`, `task-create`,
|
|
6
|
+
`worker-start`, `send`) were verified to require a live Orca terminal pane and cannot be called from the loop's
|
|
7
|
+
headless `tick`/`deliver` — every angle was tried (worker-side sends, coordinator-of-own-Run) and all failed
|
|
8
|
+
structurally, not from a config gap. What *is* reachable headlessly turned into real fixes below.
|
|
9
|
+
|
|
10
|
+
- **Escalations now carry the worker's own diagnosis**: `loop deliver`'s stuck/blocked escalations capture the
|
|
11
|
+
dispatched worker's terminal output (`terminal read --screen`, best-effort) and fold it into the Linear comment.
|
|
12
|
+
Two real incidents showed the worker had already explained the blocker in plain language (an explicit "BLOCKED:
|
|
13
|
+
..." reply, a sandboxed git error) that the idle/no-PR heuristic was discarding.
|
|
14
|
+
- **Slot assessment uses Orca's real memory diagnostic**: the harness's own `vm_stat`-based free-RAM estimate and
|
|
15
|
+
static `machine.agentRssMb` guess were measured to run roughly 2x more conservative than `orca diagnostics
|
|
16
|
+
memory`'s macOS memory-pressure reading, at the same instant. `assessSlots` now prefers `host.availableMemory`
|
|
17
|
+
and the average of real per-session RSS when Orca's diagnostic is available, falling back to the previous
|
|
18
|
+
behavior otherwise.
|
|
19
|
+
- **`ak-harness-loop` Orca Skill**: an installable `SKILL.md` (`skills/ak-harness-loop/`) documenting the
|
|
20
|
+
read-only vs. mutating `loop` commands and the operational constraints this release's investigation surfaced,
|
|
21
|
+
so a future agent session doesn't have to rediscover them.
|
|
22
|
+
|
|
3
23
|
## [0.11.0] - 2026-09-13
|
|
4
24
|
|
|
5
25
|
- **Read-only loop observability**: added `ak-harness loop observe` plus the public `runObservability` and
|