Prompt Auditor

Overview

The Prompt Auditor acts as a rigorous quality assurance mechanism for an entire ecosystem of AI agents. Its primary function is to prevent 'prompt entropy' by systematically reviewing the documentation and operational files (AGENTS.md, HEARTBEAT.md, LESSONS.md) of designated agents against a master skill template.

This agent does not rewrite code or prompts directly; instead, it analyzes performance metrics and adherence to best practices, filing structured 'Forge Issues' with concrete, proposed edits only when an agent falls below defined quality thresholds (e.g., scoring below 70/100).

Capabilities

• Systematic Auditing: Reviews a fixed rotation of agents daily (4 per run) to ensure comprehensive coverage.
• Scoring Mechanism: Computes a detailed numeric score (0-100) for each audited agent across ten predefined rubric axes.
• Issue Generation: Files precise Forge issues detailing deficiencies, citing specific evidence from the source files and proposing actionable edits.
• State Management: Tracks audit progress using LAST_AUDITED.json to manage rotation pointers and avoid redundant work.
• Reporting: Generates a daily digest summarizing scores, status, and trend analysis for stakeholders (reporting to Forge COO).

Example Use Cases

• Maintaining Agent Integrity: Run this agent nightly to ensure that all deployed agents remain up-to-date with the latest best practices without manual intervention.
• Onboarding New Agents: Before an agent is fully released, run the Prompt Auditor against its initial set of files to preemptively identify structural weaknesses.
• Compliance Checks: Use it as a scheduled task to verify that all agents adhere to mandatory security or operational guidelines embedded in the master template.