Compliance Check

Compliance Check

This is the gate. Every draft (caption, text on image, and visual concept) passes through here before it goes anywhere near Mai Marketing Machine. A post that embarrasses the owner with their carrier or their state regulator costs more than a month of content earns. When in doubt, fail it and ask.

Step 1: Load the law

Read brain/compliance/compliance-knowledge-base.md in full. That file is the only source of rules. Do not invent rules that are not written there, and do not waive rules that are. If the draft raises a question the file does not answer, stop and ask the owner, then write the answer into the knowledge base before ruling.

Step 2: Scan the complete draft

Check all three surfaces:

1. The caption, line by line
2. The text on the image (hooks on visuals are the most-seen words in the post)
3. The visual concept itself (what the image implies: fear-mongering, a competitor's branding, a person who is not the owner)

Step 3: Test against every rule category

Walk the knowledge base top to bottom. At minimum:

• Absolute claims about a stranger's coverage or price ("you are overpaying" must read "you might be overpaying")
• Poaching language ("switch to us" must read like "give us a call and we will help you review")
• Guarantees, superlatives, "always," "never," "best," "cheapest"
• Pricing, premiums, percentages, or savings amounts
• Competitor or carrier named negatively, or implied identifiably
• Carrier-specific rules (section 2), state and licensing rules (section 3), agency red lines (section 4)
• Banned phrases log (section 6): exact and near matches
• Stories: real, names removed. Any personal story must trace to brain/voice/story-bank.md. If that file does not exist, no personal story should appear; an invented anecdote is a FAIL
• Client language used where the brain calls for it

Step 4: Rule and report

Produce a verdict in this format and record it in the draft's post.md under Compliance record:

VERDICT: PASS | FAIL
Checked: caption, image text, visual concept
Rules applied: [sections of the knowledge base that were relevant]

If FAIL, for each finding:
- Location: [the exact sentence or image text]
- Concern: [the specific rule it breaks, quoted from the knowledge base]
- Suggested rewrite: [compliant alternative that keeps the hook strength]

Step 5: Route the result

• PASS: stamp post.md with Compliance: PASS (date). The draft may proceed to output/drafts/ and the posting-mmm skill.
• FAIL: send the draft back to the skill that created it with the exact concerns and suggested rewrites. Never silently rewrite and pass your own fix in the same breath: the lane skill revises, and the revision comes back through this gate. Only a PASS proceeds. There is no third outcome.

Step 6: Learn

When the owner overrules or refines a ruling, or catches something this check missed, append the lesson to the approved-phrases bank or the banned-phrases log in the knowledge base the same day. The gate gets smarter or it gets useless.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: AiAi-Mastermind
• Source: AiAi-Mastermind/aiai-mastermind-tools-and-skills
• License: MIT

Install and usage instructions live in the source repository linked above.