blast radius estimator

The problem

Autonomous agents often cause catastrophic damage during ordinary operations by deleting the wrong directories, messaging entire customer lists, or revoking critical permissions. Developers lack a systematic way to score the risk of an agent's plan before the action executes.

What it does

• Enforces a pre-action scoring protocol across five dimensions: Scope, Irreversibility, Affected Parties, Cost, and Detectability.
• Maps every planned state-change to a confirmation tier: Green (proceed), Yellow (announce), Orange (confirm), or Red (hard stop).
• Automatically escalates risk for ambiguous targets like wildcards, fuzzy matches, or stale session context.
• Provides a 'Reversibility Ladder' to transform high-risk Red actions into safe, staged pilot tests.
• Logs every estimate to create an audit trail for incident reconstruction and agent calibration.

Why this beats prompting it yourself

Simple safety prompts are easily bypassed or ignored by agents mid-execution. This skill establishes a rigid scoring discipline where unknown variables default to maximum risk, preventing the agent from 'vibing' its way through a destructive command.

Use cases

• Preventing bulk data deletion by enforcing a hard stop on pattern-matched file operations.
• Gating financial transactions or refunds behind an explicit confirmation tier based on dollar value.
• Safeguarding production environments by scoring the scope of configuration changes before they propagate.
• Validating external communications to ensure agents don't email thousands of users without a manual review.

Known limitations

Requires the agent to have visibility into the state of backups or target list sizes to avoid defaulting every action to a Red tier.