steering-safety

Enforce safety and containment rules for vault-based workflows.

8|1|Updated Apr 3, 2026
One-click install
npx skills add https://github.com/agentflock/myna --skill steering-safety
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: steering-safety
Source: https://github.com/agentflock/myna/tree/main/skills/steering-safety
Command: npx skills add https://github.com/agentflock/myna --skill steering-safety

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Myna workflows can perform actions that might expose data or violate containment rules. This Skill provides guardrails to ensure drafts are used, vault writes are restricted, and external content is treated as data only.

Core Features & Use Cases

  • Draft, never-send policy: all outbound content is prepared for user review before transmission.
  • Vault-only writes: all writes target the configured vault path, never external locations.
  • External content as data: external content is wrapped as data for processing and never executed.
  • Calendar event protection: multi-layer checks prevent unauthorized calendar changes; explicit confirmation required for writes.
  • Skill isolation: completes one task at a time without chaining other skills.

Quick Start

Enable this skill and follow its prompts to ensure drafts, vault writes, and external content are handled with explicit user confirmation.

Frequently Asked Questions about steering-safety

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent AI workflows from writing to unintended file locations?

Vault writes are restricted to a configured vault path, preventing workflows from writing to external or unintended locations. All write operations require explicit user confirmation before execution.

Can I review outbound content before it gets sent by automated workflows?

Yes, a draft, never-send policy ensures all outbound content is prepared as a draft for your explicit review. Nothing is transmitted automatically without your confirmation.

Does external content get executed when processed by containment workflows?

No, external content is treated strictly as data. It is wrapped and framed for processing only, ensuring it is never executed or triggered as code within the workflow.

How do I protect calendar events from unauthorized modifications?

Calendar event protection uses a three-layer checking system to prevent unauthorized changes. Explicit user confirmation is required for any calendar writes or bulk event operations.

Can I chain multiple skills together to run alongside safety containment?

No, skill isolation ensures the workflow completes one task at a time without chaining other skills. This prevents unintended actions from cascading across automated processes.

What is the best way to handle bulk operations safely in vault-based workflows?

Bulk operations require explicit user confirmation before proceeding. The safety containment rules apply drafting, vault-only writes, and external data framing to all bulk workflow actions.