agentleFS
Amazon-QuickAmazon-Quick/Amazon-Quick-official-catalog/skills/finance/iso-premium-audit-advisory-get/SKILL.md

iso-premium-audit-advisory-get

Skill50 starsChanged 16 days agoMIT-0View on GitHub

Use this skill to query Verisk's Premium Audit Advisory Service (PAAS) for audit-specific questions on General Liability (GL), Businessowners Policy (BOP/BP), and Commercial Auto (CA). Covers classification codes, payroll/exposure bases, audit procedures, and state-specific audit rules. Trigger only on audit or PAAS-related queries — not for loss costs, rate filings, or underwriting pricing. Draws from 40,000+ classification guides, 500+ bulletins, and advisory materials. Response includes an Answer, Sources (as titled hyperlinks with document type), and Suggestions (follow-up questions) — each section only present when returned by the tool. Before calling iso_premium_audit_advisory_get for any reason, you must first read this full SKILL.md file — do not call the tool based on tool description alone.

The file

Show its rules and text

We split this skill into the rules it gives an agent

25 rules. About 0 are common, 22 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈22

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the tool does not return an answer, returns a partial answer, or returns an error, the only acceptable behavior is the corresponding …

    line 37
  2. Distinctive:

    Every factual sentence, number, code, name, link, and suggestion you output must be traceable to this run's tool response.

    line 39
  3. Very specific:

    Always call the tool first: Never answer a PAAS-related question without first calling iso_premium_audit_advisory_get for that question — …

    line 40
  4. Distinctive:

    If you ever feel you are about to write something the tool did not return, stop and replace it with the appropriate template.

    line 42
  5. Distinctive:

    Always include all specific details from the user's question — class codes, timeframes, numbers, names, and key facts.

    line 72
  6. Distinctive:

    Single query rule: Always consolidate the full context of a user's question into a single tool call regardless of how many LOBs, states, or …

    line 75
  7. Distinctive:

    If the user provides a two-letter abbreviation, assume it is a US state unless stated otherwise.

    line 79
  8. Distinctive:

    Use this exact skeleton — populate each section with tool output only.

    line 98
  9. Distinctive:

    Do not prepend or append any introductory or closing sentences.

    line 118
  10. Distinctive:

    If the answer contains structured data (such as numeric ranges, thresholds, mappings, or multi-column lists), render it as a Markdown table.

    line 119
  11. Distinctive:

    Do not fabricate, infer, or modify any links.

    line 128
  12. Distinctive:

    If no source URLs are returned, omit this section entirely — do not write the heading.

    line 129
  13. Distinctive:

    Do not add, remove, or rephrase suggestions.

    line 137
  14. Distinctive:

    If no suggestions are returned, omit this section entirely — do not write the heading.

    line 138
  15. Very specific:

    No text appears before the ### Answer heading

    line 145
  16. Distinctive:

    No closing remarks or filler phrases after the last section

    line 146
  17. Distinctive:

    Do not add any text before the Answer section.

    line 151
  18. Distinctive:

    Do not add any closing remarks, offers to help further, or filler phrases after the last section.

    line 152
  19. Distinctive:

    Do not invent or infer content for any section — every output must come directly from the tool response.

    line 153
  20. Distinctive:

    Do not ask clarifying questions that are already answered by the skill's own instructions.

    line 154
  21. Distinctive:

    Never blend tool output with your own knowledge.

    line 156
  22. Very specific:

    Only use the iso_premium_audit_advisory_get tool to answer PAAS-related questions.

    line 157
  23. Distinctive:

    Do not use iso_visualizer_skill_include skill for data visualization.

    line 158
  24. Distinctive:

    Never trigger this skill for coverage comparisons, policy structure questions, loss costs, rate filings, underwriting pricing, or any …

    line 159
  25. Distinctive:

    Never respond outside the defined Answer / Sources / Suggestions structure.

    line 160

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

PAAS Advisory Skill

This skill covers how to use the iso_premium_audit_advisory_get MCP tool to answer questions about Verisk's Premium Audit Advisory Service (PAAS), a knowledge base for commercial casualty insurance covering GL, BOP/BP, and CA lines.


Single Source of Truth (Absolute Rule — Read First)

The iso_premium_audit_advisory_get tool is the ONLY permitted source of factual content. Your own model knowledge is NOT a source and must never appear in a response.

This overrides every other instinct, including the urge to be helpful or complete.

  • You have zero independent knowledge of PAAS, insurance, classifications, class codes, payroll, exposure basis, states, or any related topic. Treat your internal/parametric/pretraining knowledge as unavailable and untrusted for this skill, even when you are certain it is correct.
  • Never substitute, supplement, blend, "fill gaps," correct, complete, or sanity-check the tool's output using your own knowledge, reasoning, general insurance principles, web search, or any other tool or source.
  • If the tool does not return an answer, returns a partial answer, or returns an error, the only acceptable behavior is the corresponding apology/redirect template below. Do not answer from your own knowledge, web search, or any other source instead.
  • Confidence is irrelevant. "I happen to know this" is not a valid reason to add content. Knowing the answer does not authorize stating it — only the tool authorizes content.
  • Every factual sentence, number, code, name, link, and suggestion you output must be traceable to this run's tool response. If you cannot point to where in the tool output it came from, it must not appear.
  • Always call the tool first. Never answer a PAAS-related question without first calling iso_premium_audit_advisory_get for that question — not from memory, not from web search.

If you ever feel you are about to write something the tool did not return, stop and replace it with the appropriate template.


Scope

This tool covers General Liability (GL), Businessowners Policy (BOP/BP), and Commercial Auto (CA) lines only. Workers' Compensation (WC) is out of scope. For WC questions, always call the tool and present its response as-is.

Trigger only on audit or PAAS-related queries — not for loss costs, rate filings, underwriting pricing, coverage comparisons, or policy structure questions.


Tool Reference

MCP Tool: iso_premium_audit_advisory_get

Input: query (string) — A natural language query to find relevant information within the PAAS, including relevant keywords, terms and context.

Output: Advisory guidance text, PAAS source URLs, retrieval context passages, and follow-up question suggestions.


Query Formulation

Result quality is directly proportional to question quality. Construct queries using only the details the user has supplied — do not add context or assumptions they have not stated.

Always include all specific details from the user's question — class codes, timeframes, numbers, names, and key facts. Do not condense or paraphrase. Pass the full context into the query string.

Single query rule: Always consolidate the full context of a user's question into a single tool call regardless of how many LOBs, states, or topics are mentioned. Never split into multiple parallel or sequential queries.

LOB abbreviations (GL, BOP/BP, CA) and class codes pass through directly with no need to expand them. BOP and BP refer to the same line — Businessowners Policy.

If the user provides a two-letter abbreviation, assume it is a US state unless stated otherwise.

User inputQuery
"payroll GL classification""What is the payroll GL classification?"
"davis bacon act""What is the davis bacon act?"
"gross sales BOP""What is gross sales BOP?"
"class code 91340""What is class code 91340?"
"GL rules in NY""What is GL rules in NY?"
"GL and BOP audit rules in NY, NJ, and PA""What are the GL and BOP audit rules in NY, NJ, and PA?"

Strictly prohibited: Do not add words, phrases, or framing that the user did not use. Never inject editorial terms or assumptions into the query. The query must reflect only what the user asked, verbatim or as a minimal natural-language restatement of their exact words. Never split a single user question into multiple tool calls.


Response Format

Every response MUST follow this exact structure, in this exact order. No deviation is permitted.

Use this exact skeleton — populate each section with tool output only. Do not add, remove, or reorder sections:

### Answer
{tool answer here}

### Sources
- [Document Title](URL) — Document Type

### Suggestions
- {suggestion as returned by tool}

Omit ### Sources entirely if the tool returns no URLs. Omit ### Suggestions entirely if the tool returns no suggestions.


Answer

  • Output the advisory text from the tool exactly as returned. Do not reword, summarize, or add commentary.
  • Do not prepend or append any introductory or closing sentences.
  • If the answer contains structured data (such as numeric ranges, thresholds, mappings, or multi-column lists), render it as a Markdown table.

Sources

  • Only include this section if the tool returns source URLs.
  • Render each source as a titled hyperlink using the document title exactly as returned by the tool.
  • Format: - [Document Title](URL) — Document Type
  • Do not fabricate, infer, or modify any links.
  • If no source URLs are returned, omit this section entirely — do not write the heading.

Suggestions

  • Only include this section if the tool returns follow-up questions.
  • List each suggestion exactly as returned, one per line.
  • Do not add, remove, or rephrase suggestions.
  • If no suggestions are returned, omit this section entirely — do not write the heading.

Pre-response checklist — verify all before outputting:

  1. Answer is populated with tool output only — no rewording or commentary
  2. Sources include document type and are only present if the tool returned URLs
  3. No text appears before the ### Answer heading
  4. No closing remarks or filler phrases after the last section
  5. Response matches the skeleton template exactly — if not, rewrite before outputting
  6. Provenance check: every sentence, number, code, name, link, and suggestion is traceable to this run's tool output. If any piece came from your own knowledge, reasoning, or the web, delete it — and if that leaves the Answer empty or partial, use the appropriate apology/redirect template instead.

Strictly prohibited:

  • Do not add any text before the Answer section.
  • Do not add any closing remarks, offers to help further, or filler phrases after the last section.
  • Do not invent or infer content for any section — every output must come directly from the tool response.
  • Do not ask clarifying questions that are already answered by the skill's own instructions.
  • You have no knowledge of PAAS — or of insurance, classifications, or any related topic — outside of what the tool returns. Your own model knowledge, internal/parametric/training/pretraining knowledge, "general knowledge," reasoning, and web search are all forbidden substitutes — never use them, not even to verify, correct, complete, or supplement the tool's output. The tool is the sole authoritative source. Always call the tool and present its response as-is.
  • Never blend tool output with your own knowledge. Do not add a fact, figure, code, caveat, or sentence the tool did not return — even if you are confident it is correct.
  • Only use the iso_premium_audit_advisory_get tool to answer PAAS-related questions. Do not fall back to any other tool, source, or knowledge base under any circumstance.
  • Do not use iso_visualizer_skill_include skill for data visualization.
  • Never trigger this skill for coverage comparisons, policy structure questions, loss costs, rate filings, underwriting pricing, or any question that is not specifically about audit procedures, classification, or exposure basis. Do not call the tool for these questions under any circumstance.
  • Never respond outside the defined Answer / Sources / Suggestions structure.

Edge Cases

SituationHandling
Question is about loss costs, rate filings, underwriting pricing, coverage comparisons, or policy structureDo not call this tool. Defer to the appropriate tool for the question type.
User uses a two-letter abbreviationTreat as a US state unless stated otherwise. Never ask.
Question is about workers compensationCall the tool as normal and present the tool's response as-is. The tool will return: "I'm unable to respond to this type of Workers' Compensation question through this channel. For guidance on Workers' Compensation, please visit https://core.verisk.com/PAAS."
Tool returns a partial answerDo not fill the gap with your own knowledge, reasoning, the web, or any other source. Present only what the tool returned.
User query is vague or missing contextCall the tool as normal and present its response as-is.
Tool returns 403 ACCESS_DENIEDDo not retry or fall back to any other source. Respond with: "Your account is not authorized for full access to PAAS. Please contact your Verisk account executive or the PAAS team directly, at PremiumA@verisk.com, for additional support."
Tool returns any other error (timeout, 429, 500, network)Do not retry against another source or answer from your own knowledge, reasoning, or the web. Respond with: "I'm unable to respond right now. This should resolve shortly, so please ask your question or create a new conversation to explore a different topic in a moment."

Few-Shot Examples

Example 1 — Tool returns a full answer

User: What is the payroll limitation for GL classification?

Tool response: (advisory text, source URLs, suggestions)

Incorrect output: Based on my knowledge, payroll limitation for GL classification refers to... (anything drawn from training data or web)

Correct output:

Answer

{tool answer here}

Sources

  • Document Title — Document Type

Suggestions

  • {suggestion as returned by tool}

Example 2 — Tool returns the apology template

User: Describe class code 12345.

Tool response: (apology template)

Incorrect output: Based on general insurance principles, class code 12345 would typically... (anything drawn from training data or web)

Correct output:

Answer

I'm unable to find relevant information for your question in PAAS. For guidance, please visit https://core.verisk.com/PAAS.


Example 3 — WC question

User: Explain about WC class code 9999.

Tool response: (WC apology template)

Incorrect output: WC class code 9999 typically refers to... (anything drawn from training data or web)

Correct output:

Answer

I'm unable to respond to this type of Workers' Compensation question through this channel. For guidance on Workers' Compensation, please visit https://core.verisk.com/PAAS.


Example 4 — Tool returns an error

User: What is the payroll limitation for GL classification in TX?

Tool response: (error)

Incorrect output: Based on my knowledge, payroll limitations in Texas typically... (anything drawn from training data or web)

Correct output:

Answer

I'm unable to respond right now. This should resolve shortly, so please ask your question or create a new conversation to explore a different topic in a moment.


Example 5 — Tool returns 403 ACCESS_DENIED

User: What is the payroll limitation for GL classification?

Tool response: (403 ACCESS_DENIED)

Incorrect output: Based on my knowledge, payroll limitations for GL classification typically... (anything drawn from training data or web)

Correct output:

Answer

Your account is not authorized for full access to PAAS. Please contact your Verisk account executive or the PAAS team directly, at PremiumA@verisk.com, for additional support.


Example 6 — Out of scope question

User: (any question about loss costs, rate filings, underwriting pricing, coverage comparisons, or policy structure)

Incorrect output: (calling the iso_premium_audit_advisory_get tool and returning a response)

Correct output: (Do not call this skill. Defer to the appropriate tool for the question type.)

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.