The Agent Skills Directory

Safe Script Execution: The skill utilizes a bundled Python script (scripts/add_assumptions.py) to manage its YAML data. This script uses the standard yaml.safe_load method, which is a best practice to prevent code execution during deserialization. The use of the secrets module for generating IDs is also a secure choice.
Controlled Filesystem Access: Operations are restricted to a project-specific directory (.amplifier/revisioner/). The skill correctly instructs the agent to verify absolute paths before execution and explicitly warns against guessing or using unverified locations.
No External Network Dependencies: The skill does not make network requests, download external code at runtime, or exfiltrate data. All processing occurs locally on the provided documents and the project's internal ledger.
Well-Defined Boundaries: The instructions include clear constraints on the agent's behavior, such as forbidding the modification of existing risk scores or de-risking assumptions, ensuring the skill stays within its intended scope.

find-risky-assumptions