The Execution Pipeline · PreToolUse
Seven phases decide the fate of every tool call.
Each call enters at Phase 0 and descends through gates and scoring engines. The pipeline short-circuits the moment any phase emits a score that crosses the active level's deny threshold (strict 0.5, balanced 0.8, permissive 0.9); otherwise the final score is a weighted average across every engine that ran.
The Guard audit trail records each evaluation, all seven gate outcomes, all findings, and why evaluation stopped. Keep it locally or export it through OTEL Logs, independently of traces.
score → safe
suspicious
critical
PreToolUse
packet
$ —
at phase 0
0.00
scoring
idle — press ▶ replay
awaiting evaluation
no scenario running
idle