Show a recorded demo and a diagram of how it works

README.md

@@ -10,6 +10,8 @@ Reflex Control decides whether an AI task can finish on its own, retry, or escal
 > [!NOTE]
 > **Status:** Research prototype (v0.4.0). Tested on synthetic fixtures; not production-validated.
 
+![Accept, Retry and Escalate decisions for four task results](docs/demo/runtime-gate.gif)
+
 - Runs fast local checks (exit codes, test results, git diffs) alongside semantic risk signals.
 - Blocks high-risk actions like credential leaks or schema changes with a mandatory safety veto.
 - Tunes acceptance thresholds across labeled tasks with statistical confidence intervals.
@@ -96,6 +98,22 @@ Reflex Control evaluates agent execution steps through a multi-layer decision pi
 - Safe steps pass to a calibrated scoring gate that evaluates composite task quality against tuned thresholds.
 - Steps meeting the quality threshold finish autonomously (`Accept` or `Terminate`), avoiding unnecessary calls to larger models.
 
+```mermaid
+flowchart LR
+    DE["DeterministicEvidence"] --> EV["EvidenceVector"]
+    SE["SemanticEvidence"] --> EV
+    EV --> GHC["GuardedHybridComposer"]
+    GHC --> Veto["HARD SAFETY VETO RULES"]
+    GHC --> RetryRecovery["Transient Error Recovery"]
+    GHC --> RetryExhausted["Retry budget exhausted"]
+    GHC --> Quality["compute_clean_quality_index"]
+    Veto --> Escalate["Escalate"]
+    RetryRecovery --> Retry["Retry"]
+    RetryExhausted --> Escalate
+    Quality --> Accept["Accept"]
+    Quality --> Escalate
+```
+
 ---
 
 ## Workspace Architecture

docs/demo/runtime-gate.tape

@@ -0,0 +1,16 @@
+Output docs/demo/runtime-gate.gif
+Set Shell "bash"
+Set FontSize 16
+Set Width 1200
+Set Height 260
+Set Padding 24
+Set Theme "Catppuccin Mocha"
+Set TypingSpeed 40ms
+Hide
+Type `export PS1='$ ' PATH="$PWD/target/release:$PATH"` Enter
+Type "clear" Enter
+Show
+Type `example-agent-runtime-gate`
+Sleep 500ms
+Enter
+Sleep 4s