Guardrails

Safety boundaries that cannot be disabled.

---

What Are Guardrails?

Non-negotiable safety rules. Unlike permissions, boundaries cannot be overridden.

---

Core Rules

Rule	Description
Safe Paths	Only write to jaan-to/ by default
Preview First	Always show content before writing
Approval Required	User must confirm every write
No Secrets	Scan output for credentials

---

Safe Paths

Default allowed:

• jaan-to/ - All outputs
• jaan-to/outputs/ - Skill outputs

Always denied:

• src/, lib/ - Source code
• .env, *.config.* - Configuration
• Hidden directories (except jaan-to/)
• package.json - Package files

---

Preview Requirement

Before any write, you see:

Preview:
---
[content here]
---
Write this file? [y/n]

You must explicitly approve.

---

Extending Safe Paths

Add to jaan-to/context/config.md:

## Trust Overrides
- safe_paths: ["jaan-to/", "docs/"]

This allows writing to docs/ in addition to jaan-to/.

---

Why Guardrails Exist

Skills generate outputs, not production code. Guardrails ensure:

• No accidental source changes
• No credential exposure
• Human review of all outputs

---

Guardrail File

jaan-to/context/boundaries.md

Contains the detailed rules and enforcement logic.