# Money Agent Safety Harness

> A boot-time safety check ran under 242 passing tests and still let a withdrawal-enabled key through, because an absent field silently read as 'safe.' Six layers, each with mandatory evidence, including the layer most gates skip: can the agent lose the money to itself even with a perfectly permissioned key.

- Skill: `rikocr8orh8/money-agent-safety-harness` (Agent Skill)
- Install (CLI): `npx skillmds@latest add rikocr8orh8/money-agent-safety-harness`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rikocr8orh8/money-agent-safety-harness/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: rikocr8orh8 (https://skillmd.com/u/rikocr8orh8)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rikocr8orh8/money-agent-safety-harness

---


# Money-Touching Agent Safety Harness — $39

A boot-time safety check ran under 242 passing tests and still let a withdrawal-enabled key through, because an absent field silently read as 'safe.' Six layers, each with mandatory evidence, including the layer most gates skip: can the agent lose the money to itself even with a perfectly permissioned key. There is no override, on purpose.

**This is a preview.** The full skill (method, Iron Rules, dated snapshots, worked examples) is available on Agensi:
[agensi.io/skills/money-agent-safety-harness](https://www.agensi.io/skills/money-agent-safety-harness)