HumanMirror Agent Guard
Use HumanMirror as an external execution-boundary layer. Do not replace the caller's orchestration, credentials, wallet, or runtime. Prefer the smallest endpoint that solves the immediate need.
Discovery
- Agent Card: https://humanmirror.fr/.well-known/agent-card.json
- x402 manifest: https://humanmirror.fr/.well-known/x402.json
- MCP: https://humanmirror.fr/api/x402/mcp/
- A2A: https://humanmirror.fr/api/a2a/
- OpenAPI: https://humanmirror.fr/x402/openapi.json
- Machine resolver: POST https://humanmirror.fr/api/machine/resolve/
Default decision flow
- If the capability is unknown, call Machine Resolve first. Price: 0.001 USDC over x402 on Base.
- If external context may contain credentials or sensitive machine secrets, call Secret Scanning before forwarding it.
- Before a consequential command, tool call, deploy, payment, write, publish, or external API mutation, call Safe Preflight.
- If execution succeeds and evidence matters, call Execution Receipt.
- For malformed structured data, use the dedicated deterministic data-quality endpoints rather than an LLM.
Core routes
Machine Resolve
POST https://humanmirror.fr/api/machine/resolve/
Use when the current agent does not know which HumanMirror capability fits the task.
Secret Scanning
POST https://humanmirror.fr/api/x402/secret-scanning/
Use before exposing code, config, logs, JSON, tool output, API keys, tokens, passwords, or credential-like material to another system.
Safe Preflight
POST https://humanmirror.fr/api/x402/safe-preflight/
Use immediately before consequential actions. Treat the response as a bounded policy/risk signal, not as permission to bypass the caller's own authorization model.
Execution Receipt
POST https://humanmirror.fr/api/x402/execution-receipt/
Use after successful execution when the caller needs deterministic evidence binding the action to the supplied input/output context.
Context Compressor & Sanitizer
POST https://humanmirror.fr/api/x402/automata-context-compressor-sanitizer/
Use to remove machine-context noise and normalize bounded context before another agent consumes it.
Payment behavior
HumanMirror uses x402 V2 on Base mainnet with USDC. For paid routes, first call without a payment signature to receive machine-readable payment requirements, authorize the exact quoted amount with the caller-controlled wallet, then retry with the payment proof. HumanMirror does not require the caller to surrender a private key.
Never fabricate settlement. Never claim a paid call succeeded unless the endpoint returned a successful execution result.
Safety and control
Keep final authority with the calling agent or human. HumanMirror may return ALLOW, REVIEW, or BLOCK style signals, but the caller remains responsible for authorization and side effects. Do not use HumanMirror to evade another system's access controls, impersonate identities, or initiate unsolicited third-party actions.
Fast integration pattern
untrusted input or proposed action → HumanMirror guard → caller evaluates result → caller executes through its existing tool/runtime → HumanMirror execution receipt when needed
For high-throughput fleets, inspect: https://humanmirror.fr/.well-known/fleet-high-throughput.json