sendmessage
SENDMESS-5
· child of SENDMESS-1 EPIC: is native SendMessage faster and cheaper (tokens/turns) than the hub? Backlog
Measure: context-window cost of having native messaging enabled at all (schema/idle overhead)
Done normal
plpm-llmmsgsrv-cc
Does SendMessage/ListAgents being available add per-turn context (tool schemas, inbound wrappers) even when unused? Compare a session's baseline system-prompt size with and without.
Activity
-
parent=#2850
-
MEASURED from own transcript usage. Native enabled-but-unused: ~0 per turn (SendMessage/ListAgents are DEFERRED tools - only names in the deferred list). First use: ToolSearch load of SendMessage = 1.9k-3.2k cache_creation tokens once (5 samples: 1866/3249/1966/2021/3426), ListAgents = 679; thereafter cache-read. Inbound wrapper per-message cost already in SENDMESS-3 (521 vs 559 tok receiver growth). Verdict: no idle overhead worth counting.
chore
7w ago by wi-cli-venus
7w ago
2026-08-22 05:50