Update to Propalt

{
  "name": "audience-builder",
  "description": "Use when an estate or letting agent wants to build or refine a UK property prospecting list, vendor or landlord audience, direct-mail target list, or a filtered set of properties to contact. Use property and market signals from Propalt; do not use this skill for a simple already-specified property query that only needs execution, or for targeting people based on inferred sensitive or household characteristics.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 434
    }
  ],
  "skill_md_contents": "---\nname: audience-builder\ndescription: \"Use when an estate or letting agent wants to build or refine a UK property prospecting list, vendor or landlord audience, direct-mail target list, or a filtered set of properties to contact. Use property and market signals from Propalt; do not use this skill for a simple already-specified property query that only needs execution, or for targeting people based on inferred sensitive or household characteristics.\"\n---\n\n# Audience Builder\n\nBuild a defensible property prospecting audience from Propalt data. The user supplies the business goal and patch; Propalt supplies property and market evidence.\n\n## Workflow\n\n1. Establish the patch and goal. Prefer an outward postcode, supported sectors, or a user-supplied polygon. If the user gives only vague place text, use `street_lookup` or `address_lookup` to resolve it before continuing.\n2. Sense-check the proposed audience with local property-market data. Use only calls relevant to the hypothesis, typically `get_market_analysis`, `get_monthly_market`, and `get_quarterly_market`.\n3. Build or count the audience with the appropriate search:\n   - residential property filters: `search_properties`;\n   - live/recent sales signals: `selling_property_market_activity_properties_for_sale_post`;\n   - live/recent lettings signals: `letting_property_market_activity_properties_for_let_post`;\n   - explicit landlord/proprietor prospecting: `landlords_organisation_landlords_post`.\n4. Start with `return_type=\"count\"` when the user is still refining criteria. Retrieve records only after the criteria are useful and the user wants examples or a list.\n5. Present the audience size, the filters used, why the signal is relevant in this patch, and a concise contact/message angle grounded in the same evidence.\n6. If the user wants the resulting property list exported for Propalt My Lists, hand off conceptually to the `propalt-csv-export` workflow.\n\n## Guardrails\n\n- Do not call full-postcode-only tools such as `get_epc_fabric_by_postcode`, `get_demographics`, `get_households`, or `get_income` with an outward postcode.\n- For EPC targeting at outward-postcode level, use supported EPC filters in `search_properties` or the relevant market-activity search rather than pretending `get_epc_fabric_by_postcode` supports an outcode.\n- Aggregate demographics can be used only as area context when a valid full postcode, LSOA, or local-authority identifier is available. Do not infer the age, income, family status, vulnerability, or other personal characteristics of a particular household from area statistics, and do not use those inferred characteristics to select individual homes for contact.\n- Do not manufacture urgency, likely-to-sell claims, or personal circumstances that Propalt does not provide.\n- If the user gives a precise supported filter set and only wants it executed, run the appropriate search directly without forcing a long strategy workshop.\n\n## Output\n\nReturn: audience label, exact search criteria, count, 3–10 representative records if requested, local evidence that supports the angle, limitations, and a short outreach hook. Keep property facts separate from inferred marketing strategy.\n"
}