overview

This skill establishes the core mental model for TraceMem, positioning the AI agent as a transparent participant within a strictly governed execution environment. It defines the necessary framework for agents operating in sensitive contexts—such as reading private data, modifying system states, or processing financial transactions—where accountability and auditability are non-negotiable. By treating the environment as a courtroom, the agent learns to present evidence through decision reads, justify actions via policies, and maintain an immutable audit trail of its reasoning and outcomes.

The overview covers the critical distinction between the mandatory Decision Envelope—the container for all governed actions—and lightweight discovery or recording tasks. It guides agents through the lifecycle of an action, from the initiation of a decision_id to the final commitment or abortion of a process. This approach helps prevent common pitfalls, such as bypassing governance, attempting operations without authorization, or neglecting the requirement for human-in-the-loop approvals.

• Foundational Principles: Understands the Courtroom model, where every action must be supported by evidence, policy citations, and explicit logic logging.

• Decision Envelope Management: Strict enforcement of the mandatory decision_create, decision_read, decision_write, and decision_close workflow for any data-sensitive operation.

• Governance Integration: Explains how to interface with decision_evaluate and decision_request_approval to respect system boundaries and human oversight.

• Immutable Audit Trails: Ensures that all agent operations are recorded for future auditing, emphasizing the prohibition of secrets or PII in non-secure fields.

• Operational Clarity: Distinguishes when to use the full Decision Envelope workflow versus one-shot tools like decision_record.

• Mandatory Usage: All governed data operations (read/write/evaluate) require an active decision_id; attempting these without one will trigger failures.

• Immutable Logs: Never store API keys, passwords, or PII in intent strings or context summaries; these logs are permanent.

• Automation Awareness: Agents must explicitly map their activity to authorized automation modes (propose, approve, override, autonomous) to ensure intent matches policy.

• Discovery Precedence: Use products_list and product_get for information gathering before opening a decision envelope to keep trace logs clean and focused on action.

• Error Handling: Always handle decision_request_approval rejection by aborting the current operation rather than bypassing governance.