I run a dozen AI agents. I just audited all of them with one prompt.
Most people's AI setup is a junk drawer. Mine was too - a dozen agents and prompts scattered across two computers, built at different times, drifting in different directions. This week I audited the whole system with one prompt. Here's what it caught, and the prompt itself, free.
Some background: I use AI as an extension of the business. Agents draft content in my voice, compile nightly digests of what shipped, capture receipts, brief me on the day's marketing calendar, and track finances. Each one worked on the day I built it. The problem is what happens after - the business changes, the positioning sharpens, and the agents keep running on last quarter's instructions.
What the audit caught
I ran every agent, skill, and prompt through the audit below. Four findings paid for the afternoon:
- An agent that only worked when one laptop was open. My content pipeline lived on a specific machine as a scheduled desktop task. If the laptop slept, the pipeline slept. The audit's fix: rebuild it as a cloud routine that runs on a schedule no matter what my hardware is doing.
- A full strategy for something I'd already killed. A detailed Instagram playbook was still sitting in my system for an account I had shut down. Dead strategy is worse than no strategy - the next tool that reads it treats it as truth. Archived it, salvaged the 30% that still applied.
- My own banned words, inside my own prompts. I keep a strict voice guard - words and constructions my content never uses. The audit found my old positioning language living inside a scoring prompt I wrote months ago. Every draft that prompt produced was quietly pulling against my current positioning.
- Three "master plans" that disagreed. Different documents, written in different tools at different times, each claiming to be the plan. The audit forced one source of truth and demoted the rest to archives.
None of these were dramatic failures. That's the point. AI systems don't usually break loudly - they drift quietly, and the outputs get a little more generic, a little more off-brand, a little more wrong every week. An audit is how you catch drift before your audience does.
AI systems don't break loudly. They drift quietly.
The prompt
Paste this into your AI tool of choice, then feed it your agents, skills, prompts, and workflows. It works on one prompt or fifty. Copy it whole:
You are operating as an elite AI systems architect, agent designer, workflow strategist, and prompt engineer. Review, diagnose, improve, and standardize all of the AI skills, agents, prompts, automations, and workflows I provide. Cleaner prompts are not the goal. The goal is an AI operating system that is sharper, safer, more reliable, more autonomous, and more commercially useful. Context: I use AI as an extension of my business and brain. My work spans marketing, brand strategy, retail operations, PR, email/SMS, website content, productized consulting resources, personal brand development, calendar and email workflows, business documentation, creative direction, and launch planning. Your job: audit everything I provide and improve it as if you were preparing it for a high-performing founder, strategist, and operator who needs reliable outputs with minimal back-and-forth. Review each item for: 1. Strategic clarity - is the purpose obvious? Is it solving the right problem at the right level of ambition? 2. Output quality - will the output be specific, polished, structured, and ready to use? 3. Autonomy - does it know when to proceed, when to ask, and when to flag uncertainty? 4. Tool and context use - does it request the right files, data, brand context, or research when needed? 5. Role definition - is the expertise specific enough? Does it behave like a senior operator, strategist, creative director, analyst, or reviewer depending on the task? 6. Constraints and guardrails - does it avoid hallucination, generic advice, brand drift, compliance issues, privacy issues, and unsafe assumptions? 7. Workflow design - are steps missing, redundant, or in the wrong order? 8. Brand and tone alignment - does it reflect my working style: direct, strategic, high-standard, practical, commercially aware? 9. Reusability - can it run repeatedly without breaking, going stale, or needing manual setup? 10. Commercial value - does it save time, make money, improve execution, protect the business, or produce better client-facing assets? Process: first, inventory everything I provide, grouped by category (marketing and brand, customer communications, operations, creative direction, personal brand and content, digital products, admin and assistant workflows, documentation, review agents, anything else you identify). Then run a diagnostic on each item: what it currently does, what is strong, what is weak or unclear, what risks or failure modes exist, what should be upgraded, and whether it should be kept, merged, rewritten, split, or retired. Then deliver: A. Executive summary - highest-leverage findings first: what works, what is broken, what is redundant, what needs a rebuild. B. Priority roadmap - immediate fixes, high-value rebuilds, nice-to-have refinements, items to retire or consolidate. C. Audit table - name, current purpose, main issue, recommended action, priority, expected benefit. D. Rewritten versions - for each item worth improving, a stronger replacement including: name, purpose, best use case, input required, operating instructions, decision rules, output format, quality bar, guardrails, and an example trigger command. E. System-wide standards - a reusable standard for all future agents: naming conventions, prompt structure, output formatting, when to ask vs. proceed, how to handle missing context, how to preserve brand voice, how to flag uncertain information, and how to produce finished deliverables instead of generic advice. F. Missing pieces - the agents and workflows I should have but don't, prioritized by what would help me move faster, produce better assets, sell more, improve delivery, cut repetitive work, and prevent compliance or communication mistakes. G. The final operating system - the ideal structure for my full AI workflow system, organized like an internal operating manual. Constraints: - No generic improvements. No watering down the voice or the strategy. - Do not overcomplicate prompts to make them longer. - Do not invent tools, files, or access I did not provide. - Preserve each agent's original intent unless you clearly explain why it should change. - Flag anything that needs human review. - When something is unclear, make the smartest reasonable assumption and label it. - Prioritize practical business value over theoretical prompt engineering. - Output everything in a clean, copy-paste-ready format.
How to get the most out of it
- Feed it everything, including the embarrassing stuff. The abandoned prompts and half-built agents are where the audit earns its keep.
- Run it quarterly. Drift is a function of time. My rule now: every agent gets re-audited when the positioning changes, and no less than once a quarter.
- Act on section E first. The system-wide standard is the highest-leverage output - one voice guard and one facts file that every agent reads beats fixing agents one at a time.
- In regulated categories, add your compliance rules to the constraints. I run cannabis marketing, so every agent of mine carries the same hard line: no health claims, nothing that could read under-21, state disclaimers where required. Your version of that list belongs in the prompt.
That's the self-serve version. If you'd rather have a second set of eyes on your setup - what to keep, kill, and rebuild, mapped to your business - that's the AI Systems Audit I run for founders and operators. Bring the junk drawer; you'll leave with an operating system. More on AI marketing systems and generative engine optimization.