Claude Code internally spawns multiple sub-agents — planner, file ops, executor, validator.
Every tool call arrives at the MCP proxy wearing the same badge: claude-code.
Behavry is the only layer that knows what actually happened.
0 intercepted0 blocked0 escalated
Claude Code internal
Claude Code spawns sub-agents internally — each with its own identity and intent
Behavry proxy · inline execution path
Proxy active · all MCP traffic intercepted before execution
Every tool call passes through before it fires
Decision Trace
Causal chain-of-custody artifact — not a log
The attestation gap: Behavry intercepted 0 tool calls from
4 distinct internal agents — all arriving as
claude-code. The Decision Trace above is the only artifact that proves
what each sub-agent actually did. Without inline interception, this chain cannot be reconstructed.