The Agent Skills Directory

[SAFE]: The skill implements a security-enhancing architecture (Human-in-the-loop) designed to prevent autonomous agents from communicating with external counterparties without human oversight.\n- [SAFE]: Input sanitization is explicitly defined in the 'Filing a draft' protocol, requiring the removal of control characters and enforcement of field length limits to prevent malformed data from entering the ledger.\n- [SAFE]: Data integrity and state transitions are strictly enforced at the database level in references/approval-ledger.sql using complex SQLite triggers that prevent unauthorized snapshots, immutable record tampering, and race conditions.\n- [SAFE]: The Python implementation in references/approval_claims.py adheres to high security standards, utilizing parameterized SQL queries for injection prevention, secrets.compare_digest for timing-safe hash verification, and atomic database transactions.\n- [SAFE]: Audit trails are ensured through the use of SHA-256 hashing for draft text and unique delivery receipts, providing a verifiable link between operator decisions and dispatched messages.

operator-approval-loop