Update to Press Loft

{
  "description": "Use the connected Press Loft tools for journalist image and editorial research, publication-confirmed image downloads, or brand/agency library reporting and selected product edits. Use for Press Loft requests, not unrelated product research.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 552
    }
  ],
  "name": "pressloft-library",
  "skill_md_contents": "---\nname: pressloft-library\ndescription: Use the connected Press Loft tools for journalist image and editorial research, publication-confirmed image downloads, or brand/agency library reporting and selected product edits. Use for Press Loft requests, not unrelated product research.\n---\n\nUse the tools available to the connected account and their current schemas.\nJournalist and brand/agency accounts expose different tools. Follow returned\npagination and permissions; one page is not the full result set. Country,\nlanguage and currency follow saved settings unless the user requests overrides.\n\n## Journalist images and editorial content\n\n- Use `search_images` for product/lifestyle images. Preserve localised names,\n  brands and prices. Prices describe products, not image licence fees. Search\n  returns medium previews and does not request high-resolution downloads.\n- Image search already attaches its image-card component. Use `show_image_card`\n  for a specifically requested single image. When combining searches or showing\n  additional selections, call `show_image_cards` once with every actual image ID\n  to present and the same localisation overrides. Do not call a display tool\n  per image or repeat it for the same completed selection.\n- Use `search_press_releases`, `search_trends` and `search_collections` for\n  discovery, then the matching `get_press_release`, `get_trend` or\n  `get_collection` to read selected content and its paginated `images` list.\n  Listing `preview_images` references are not the complete associated image list.\n- Present full requested press release bodies, trend descriptions/editorial\n  details and collection descriptions as plain text. Editorial content has no\n  card UI. To display its associated images, pass actual IDs from the detail\n  result's `images` list to `show_image_cards`; never use a release, trend or\n  collection ID as an image ID. Keep the editorial text alongside the image\n  cards, which reload the products' actual names and details.\n- When image cards are available, do not duplicate them with Markdown images.\n  In clients without component rendering, use only the supplied image-display\n  result's medium Markdown sources. Do not invent sources or print standalone\n  image URLs. Native image blocks alone do not establish visible rendering.\n- Use `list_brands` for the directory: `country_code` filters brand location,\n  `ships_to` filters shipping coverage, and `sort: \"name\"` requests alphabetical\n  results. Shipping defaults to the journalist's registered country;\n  `ships_to: \"ALL\"` browses worldwide. Directory discovery does not grant access\n  to private press-office content.\n\nFor an explicit high-resolution download request, let the journalist choose an\nimage, an associated publication or a custom publication name using Other, and\nan exact publication date. Use the card form when available. The\n`list_download_publications` tool only loads choices/defaults;\n`request_image_download` records a confirmed request. Never invent publication\ndetails or treat pre-filled defaults as confirmation for another image. Reuse\nthe request ID only for identical retries; changed details require a new ID.\nDo not create download requests during search or editorial browsing. Temporary\ndownload links belong to the download interaction, not search previews.\n\n## Brand and agency libraries\n\n- For headline image counts, call `get_image_counts`. Include its `as_of` time.\n  Live, hidden and incomplete counts can overlap; do not sum them into a total.\n- For products, call `list_products` with the requested filters. Respect its\n  pagination; one returned page is not the entire library or a complete count.\n- `list_categories` returns the site's fixed taxonomy, not the categories used\n  by this account. Use returned category names when filtering products, and\n  subcategory IDs when assigning categories in a product edit.\n- For download summaries, use `get_download_totals`; for individual download\n  records, use `list_downloads`. Pass the requested date range when supplied.\n- Use `list_press_releases` and `list_collections` for the connected brand's\n  published content. These are distinct from journalist discovery tools.\n- Agency listing tools include linked clients by default. Label combined\n  agency results as such. When the user asks about one brand, resolve its\n  `client_id` through `list_clients` when that tool is available and apply the\n  filter. Ask which brand is intended when returned names are ambiguous.\n\nWhen the user requests an edit, identify the specific product and requested\nfields before using `edit_product_caption`. Categories and keywords replace\ntheir complete lists, while omitted fields stay unchanged. Explain replacement\nsemantics when the user's intended lists are unclear. Report the actual saved\nresult and any live-check warning returned by the server.\n\nFor up to 50 explicitly selected images, use `bulk_edit_product_captions` for\nindividual updates or `bulk_edit_product_taxonomy` for one shared keywords and/or\ncategories replacement. Resolve the intended images and updates before writing;\ndo not silently expand a selection into library-wide loops. Reuse each\n`request_id` with the identical payload for retries. A returned `batch_id` means\nthe request was queued, not that edits succeeded. Poll `get_product_caption_batch`\nand report per-image save/live-check outcomes and warnings. If processing remains\npending, report that state without claiming completion. Follow the returned\nwarning for recovery; do not change a request ID merely because a response was\nlost. Product deletion is not supported.\n\nIf the Press Loft connection requires authentication, direct the user to connect\ntheir account through ChatGPT's app controls. Account scope and permissions are\nenforced by the server; do not treat user-supplied account IDs as authorisation.\nKeep account passwords, OAuth tokens and other sign-in secrets out of chat and\ntool arguments.\n"
}