You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- Long-horizon reasoning critic default: `moonshotai/kimi-k3` with mandatory `max` reasoning, no temperature parameter, and a receipt-recorded 4096-token critic budget.
310
311
- REGULAR smoke/simple prompts auto-route to `openrouter_single` with the GLM principal and `wsp_holo` HoloIndex grounding (no Fusion panel, Skillz, or git).
311
312
- Context is not a 012-facing selector; it is resolved from WSP_15 tier.
312
313
- Skillz/Wardrobe/Rolodex/OpenClaw/Hermes discovery is context only. RedDog may recommend a governed handoff, but this extension cannot execute it.
- Added the verified OpenRouter slug `moonshotai/kimi-k3` to the default manual Fusion panel while retaining Kimi K2.7 Code for implementation comparison and capacity fallback.
8
+
- Kimi K3 requests use its published mandatory `max` reasoning contract, omit the unsupported temperature field, and use a receipt-recorded 4096-token critic budget validated by the live compatibility smoke.
9
+
- Review packets continue to bind the exact requested panel model IDs; no mutable `kimi-latest` alias is used.
@@ -191,12 +191,13 @@ The lead is configurable. Use Cursor settings or workspace/user settings:
191
191
"reddog.leadModel": "z-ai/glm-5.2",
192
192
"reddog.panelModels": [
193
193
"deepseek/deepseek-v4-pro",
194
-
"moonshotai/kimi-k2.7-code"
194
+
"moonshotai/kimi-k2.7-code",
195
+
"moonshotai/kimi-k3"
195
196
]
196
197
}
197
198
```
198
199
199
-
The extension uses up to four panel models. RedDog defaults to GLM-5.2 as principal, DeepSeek V4 Pro as adversarial critic, and Kimi K2.7 Code as implementation critic.
200
+
The extension uses up to four panel models. RedDog defaults to GLM-5.2 as principal, DeepSeek V4 Pro as adversarial critic, Kimi K2.7 Code as implementation critic, and Kimi K3 as a long-horizon reasoning critic. Kimi K3 uses OpenRouter's explicit `moonshotai/kimi-k3` slug; its request omits unsupported temperature, records mandatory `max` reasoning, and uses a 4096-token critic budget because lower budgets did not produce quorum-usable final output in the live compatibility smoke. The review packet records the actual per-panel token budgets.
Copy file name to clipboardExpand all lines: extensions/reddog/tests/TestModLog.md
+7Lines changed: 7 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,5 +1,12 @@
1
1
# Foundups(R)Agent TestModLog
2
2
3
+
## 2026-07-18 - REDDOG_FUSION_KIMI_K3_PHASE1
4
+
5
+
- Asserted the 0.4.1 package/runtime version lock and default Kimi K3 panel membership.
6
+
- Asserted the bridge emits Kimi K3 requests with mandatory `max` reasoning, without temperature, and with the 4096-token panel budget.
7
+
- Retained Kimi K2.7 Code coverage so the default panel can compare the two Kimi generations.
8
+
- Reconciled the stale resident-session contract assertion with the already-shipped durable AgentDB runtime symbol; production resident code was not changed.
assert.strictEqual(spinePreview.governed_work_order_candidate.red_dog_instance_id,'foundups-agent-0.4.0','candidate must bind extension version');
1095
+
assert.strictEqual(spinePreview.governed_work_order_candidate.red_dog_instance_id,'foundups-agent-0.4.1','candidate must bind extension version');
1091
1096
assert.strictEqual(spinePreview.governed_work_order_candidate.repo_permission_snapshot.source,'extension_runtime_candidate','candidate must not forge permission source');
1092
1097
assert.strictEqual(spinePreview.governed_work_order_candidate.repo_permission_snapshot.permission_level,'needs_verification','candidate must fail closed on permission');
0 commit comments