create-manychat-ai-agent 0.29.0 → 0.30.0
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/agent/.env.example +7 -0
- package/agent/package.json +1 -1
- package/package.json +1 -1
package/agent/.env.example
CHANGED
|
@@ -139,3 +139,10 @@ LOG_LEVEL=info
|
|
|
139
139
|
# and delivered. A tenant on a reasoning model wants a looser figure for the
|
|
140
140
|
# suite without touching the request path.
|
|
141
141
|
# EVAL_MAX_LATENCY_MS=30000
|
|
142
|
+
#
|
|
143
|
+
# A second model that grades replies during `pnpm eval`, as provider:model
|
|
144
|
+
# (specs/016-model-graded-evals.md). Every run calibrates it first against
|
|
145
|
+
# <EVAL_DIR>/calibration.jsonl, and only the groups it agrees with unanimously
|
|
146
|
+
# grade anything. Three calls per criterion per case, plus three per calibration
|
|
147
|
+
# item. Unset: no judge, and the suite behaves as specs/009 defines.
|
|
148
|
+
# EVAL_JUDGE_MODEL=anthropic:claude-sonnet-5-5
|
package/agent/package.json
CHANGED