assumption auditor

The problem

AI agents often fail because they act on unexamined guesses about data schemas, API behaviors, or human availability. These silent failures happen when an agent assumes a "happy path" without verifying if the underlying context is still true or even exists.

What it does

• Extracts a minimum of 8 hidden assumptions from any proposed plan or answer.
• Grades every assumption based on evidence: Verified, Sourced, Inferred, or Hoped.
• Calculates the "blast radius" for load-bearing assumptions to prioritize verification efforts.
• Forces explicit definition of "done-criteria" and human-in-the-loop fallback procedures.
• Generates a persistent assumption ledger to track risk throughout the task execution.

Why this beats prompting it yourself

Standard prompting struggles to catch "silent" assumptions because agents are biased toward completion. This skill implements a rigorous auditing protocol that treats "probably fine" as a failure state, forcing the agent to prove its foundation before executing irreversible actions.

Use cases

• Pre-flight checks for complex multi-step agent workflows to prevent logic errors.
• Root-cause analysis for agents that provided "confidently wrong" answers.
• Risk assessment for agents handling destructive actions like database writes or API deletions.
• Validation of delegation contracts between a primary agent and its subagents.

Known limitations

This skill governs planning and factual assumptions, not user psychology. If a user explicitly forces an assumption, the auditor documents the risk but will not override the user's authority.