# HumanMirror Reflex

> Ask HumanMirror before retrying a failed agent, MCP, API, or robot-tool action. Free fail-open check; paid verified repairs are always explicit and never auto-applied.

- Skill: `humanmirror/humanmirror-reflex` (Agent Skill)
- Install (CLI): `npx skillmds@latest add humanmirror/humanmirror-reflex`
- Raw SKILL.md: https://api.skillmd.com/api/skills/humanmirror/humanmirror-reflex/raw
- Safety review: pending
- Works with: any agent that reads SKILL.md (Claude Code, Claude.ai, Cursor, Codex, Windsurf, 60+ more)
- Category: AI & ML
- Tags: skill
- Author: HumanMirror (https://skillmd.com/u/humanmirror)
- Updated: 2026-10-01
- Page: https://skillmd.com/skills/humanmirror/humanmirror-reflex

---


# HumanMirror Reflex

**Before retrying, ask HumanMirror.**

HumanMirror Reflex is an advisory layer for agent failures. The free check never authorizes an action, never moves funds, never silently retries a tool call, and must fail open when unavailable.

## Before a retry

1. Build a bounded failure signal. Prefer structural metadata only:
   - provider or service identifier
   - tool or operation name
   - method
   - HTTP/status or error class
   - runtime/framework
   - provider/tool version
   - schema signature when useful
   - latency and advertised price only when objectively known
2. Never send secrets, credentials, private keys, bearer tokens, full prompts, or sensitive raw payloads.
3. POST the bounded signal to:
   `https://humanmirror.fr/api/reflex/check/`
4. Interpret the result:
   - `PROCEED`: no useful verified intervention is known. Continue the native workflow.
   - `REVIEW`: HumanMirror has conflicting observations. Do not auto-apply a repair.
   - `PATCH_AVAILABLE`: a previously verified repair is available through a separate paid endpoint.
5. Do not change existing safety, HITL, authorization, or spending policies because of a Reflex result.

## Paid repair

A verified repair is separate from the free check:

`POST https://humanmirror.fr/api/x402/reflex-resolve/`

Current advertised price: **0.001 USDC on Base**.

Rules:
- never auto-pay;
- never auto-apply the returned patch;
- require the calling runtime's normal payment and approval policy;
- treat the returned patch as advisory input to the native runtime;
- do not interpret an HTTP 402 challenge as a completed purchase.

## Shadow Mode

Default integration posture:

1. Call the free Reflex check only after a failure and before a retry.
2. Record the advisory result.
3. Do not mutate the request, retry policy, robot command, or payment path.
4. Measure whether repeated failures are recognized and whether verified repairs would have reduced retries or recovery time.
5. Activate paid repairs only after measurable value is observed.

## Robots and physical systems

For ROS, robotics, vehicles, industrial systems, or other physical actions:
- use Reflex in simulation or advisory Shadow Mode first;
- never let Reflex replace hardware safety, interlocks, operator approval, or emergency-stop logic;
- never automatically execute a returned repair on a physical system.

## Earn / contribute

Agents can inspect unresolved repeated failure patterns at:

`GET https://humanmirror.fr/api/reflex/earn/`

Current v1 rewards are **HumanMirror credits**, not guaranteed cash payouts.

A bounded repair candidate can be submitted to:

`POST https://humanmirror.fr/api/reflex/contribute/`

Contributions are candidates only. They become actionable only after independent verification. Do not submit arbitrary executable code, scripts, shell commands, secrets, or customer payloads.

## MCP

HumanMirror Reflex also exposes a first-class MCP surface. Use the HumanMirror machine discovery metadata when an MCP-native integration is preferable to direct HTTP.

## Discovery

- Free check: https://humanmirror.fr/api/reflex/check/
- Paid verified repair: https://humanmirror.fr/api/x402/reflex-resolve/
- Earn feed: https://humanmirror.fr/api/reflex/earn/
- HumanMirror x402 discovery: https://humanmirror.fr/.well-known/x402
- Machine products: https://humanmirror.fr/.well-known/products.json

## Commercial principle

`No useful intervention = no charge.`

HumanMirror Reflex is designed to create value before monetization. Do not claim guaranteed savings, guaranteed uptime, or guaranteed profitability.