The Agent Skills Directory

[SAFE]: The skill implements a comprehensive security and privacy framework for verifying AI-generated goals.
[SAFE]: Includes a dedicated privacy scanning library (scripts/lib/privacy.ts) that uses robust regex patterns to identify and redact sensitive information like API keys, private keys, tokens, and PII from logs, DOM snapshots, and URLs.
[SAFE]: Enforces a single-writer lease mechanism (scripts/checkpoint-readiness.ts) using revision tracking and SHA-256 content hashes to ensure the integrity of the goal contract throughout its lifecycle.
[SAFE]: Restricts network operations for evidence collection to well-known and trusted domains, specifically whitelisting GitHub's attachment infrastructure to prevent unauthorized data exfiltration.
[SAFE]: Instructions explicitly prohibit high-risk activities, such as the use of sudo or the implementation of product code during the goal-authoring phase, maintaining a strict boundary between planning and execution.
[SAFE]: All external verifier tools are required to be installed in a non-privileged, user-scoped, and reversible manner, with mandatory health checks and capability smoke tests.

verify-goal