Previous value: -"The minimal set of files the specialist needs. Each entry is { name, url?, content_base64? } and must carry exactly one source. `url` is strongly preferred: pass a link if the user gave one (Google Drive / Dropbox / S3 / raw GitHub), or — if you have a way to upload the file yourself (e.g. a code execution / shell environment) — upload it to any URL-reachable location first and pass that link instead of the raw bytes; the server just downloads it, which is fast and costs you almost no extra tokens. Use `content_base64` only as a last resort, when there is no URL and you have no way to produce one yourself (e.g. a small file the user attached directly in a chat-only client with no upload capability of its own) — inlining bytes means writing out the entire file as text, which is slow and burns far more tokens than a URL for anything beyond a tiny file. Before falling back to it, tell the user that creating the task will take noticeably longer and use more tokens because of it. Raw base64 only, never a `data:...;base64,` prefix. This server runs remotely and cannot read the caller's disk, so never pass local paths. The server scans every file for secrets, then uploads the rest to the task's file storage on Stoory; contents never enter the model context. Max 5MB per file. Entries that are unreadable or filtered as secrets come back in blocked_files — the task is still created without them."New value: +"The minimal set of files the specialist needs. Each entry is { name, url?, content_base64? } and must carry exactly one source. `url` is strongly preferred: pass a link if the user gave one (Google Drive / Dropbox / S3 / raw GitHub) — the server rewrites common Google Drive and Dropbox share links to their direct-download form, and rejects a URL that returns a web page instead of a file — or, if you have a way to upload the file yourself (e.g. a code execution / shell environment) — upload it to any URL-reachable location first and pass that link instead of the raw bytes; the server just downloads it, which is fast and costs you almost no extra tokens. Use `content_base64` only as a last resort, when there is no URL and you have no way to produce one yourself (e.g. a small file the user attached directly in a chat-only client with no upload capability of its own) — inlining bytes means writing out the entire file as text, which is slow and burns far more tokens than a URL for anything beyond a tiny file. Before falling back to it, tell the user that creating the task will take noticeably longer and use more tokens because of it. Raw base64 only, never a `data:...;base64,` prefix. This server runs remotely and cannot read the caller's disk, so never pass local paths. The server scans every file for secrets, then uploads the rest to the task's file storage on Stoory; contents never enter the model context. Max 5MB per file. Entries that are unreadable or filtered as secrets come back in blocked_files — the task is still created without them."

Previous value: -"The minimal set of files the specialist needs. Each entry is { name, url?, content_base64? } and must carry exactly one source. `url` is strongly preferred: pass a link if the user gave one (Google Drive / Dropbox / S3 / raw GitHub), or — if you have a way to upload the file yourself (e.g. a code execution / shell environment) — upload it to any URL-reachable location first and pass that link instead of the raw bytes; the server just downloads it, which is fast and costs you almost no extra tokens. Use `content_base64` only as a last resort, when there is no URL and you have no way to produce one yourself (e.g. a small file the user attached directly in a chat-only client with no upload capability of its own) — inlining bytes means writing out the entire file as text, which is slow and burns far more tokens than a URL for anything beyond a tiny file. Before falling back to it, tell the user that creating the task will take noticeably longer and use more tokens because of it. Raw base64 only, never a `data:...;base64,` prefix. This server runs remotely and cannot read the caller's disk, so never pass local paths. For a large binary with no URL and no way to upload it yourself, don't inline it — tell the user to add it to the task repo manually instead. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. A non-empty list always gets a repo, even if every entry ends up in blocked_files (unreadable or filtered) — so the 'add it manually' fallback always has a real repo_url to point to. The GitHub username used to invite the client into the task repo comes from their Stoory account, not from you — you never see or pass it. If create_task fails because it's missing or invalid, tell the user to check/set their GitHub username in their Stoory account settings, then retry create_task."New value: +"The minimal set of files the specialist needs. Each entry is { name, url?, content_base64? } and must carry exactly one source. `url` is strongly preferred: pass a link if the user gave one (Google Drive / Dropbox / S3 / raw GitHub), or — if you have a way to upload the file yourself (e.g. a code execution / shell environment) — upload it to any URL-reachable location first and pass that link instead of the raw bytes; the server just downloads it, which is fast and costs you almost no extra tokens. Use `content_base64` only as a last resort, when there is no URL and you have no way to produce one yourself (e.g. a small file the user attached directly in a chat-only client with no upload capability of its own) — inlining bytes means writing out the entire file as text, which is slow and burns far more tokens than a URL for anything beyond a tiny file. Before falling back to it, tell the user that creating the task will take noticeably longer and use more tokens because of it. Raw base64 only, never a `data:...;base64,` prefix. This server runs remotely and cannot read the caller's disk, so never pass local paths. The server scans every file for secrets, then uploads the rest to the task's file storage on Stoory; contents never enter the model context. Max 5MB per file. Entries that are unreadable or filtered as secrets come back in blocked_files — the task is still created without them."

Previous value: -"The minimal set of files the specialist needs. This server runs remotely and cannot read the caller's disk, so do NOT pass local paths. Each entry is { name, url?, content_base64? } and must carry exactly one source: `url` (the server downloads it — prefer this when the user gave a link, e.g. Google Drive / Dropbox / S3 / raw GitHub) or `content_base64` (inline bytes, for small local or chat-uploaded files; raw base64 only, no `data:...;base64,` prefix). For a large binary with no URL (e.g. an image), do not inline it — tell the user to add it to the task repo manually. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. A non-empty list always gets a repo, even if every entry ends up in blocked_files (unreadable or filtered) — so the 'add it manually' fallback always has a real repo_url to point to. The user must have github_username set — if create_task reports it missing, ask the user for their GitHub login and try again."New value: +"The minimal set of files the specialist needs. Each entry is { name, url?, content_base64? } and must carry exactly one source. `url` is strongly preferred: pass a link if the user gave one (Google Drive / Dropbox / S3 / raw GitHub), or — if you have a way to upload the file yourself (e.g. a code execution / shell environment) — upload it to any URL-reachable location first and pass that link instead of the raw bytes; the server just downloads it, which is fast and costs you almost no extra tokens. Use `content_base64` only as a last resort, when there is no URL and you have no way to produce one yourself (e.g. a small file the user attached directly in a chat-only client with no upload capability of its own) — inlining bytes means writing out the entire file as text, which is slow and burns far more tokens than a URL for anything beyond a tiny file. Before falling back to it, tell the user that creating the task will take noticeably longer and use more tokens because of it. Raw base64 only, never a `data:...;base64,` prefix. This server runs remotely and cannot read the caller's disk, so never pass local paths. For a large binary with no URL and no way to upload it yourself, don't inline it — tell the user to add it to the task repo manually instead. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. A non-empty list always gets a repo, even if every entry ends up in blocked_files (unreadable or filtered) — so the 'add it manually' fallback always has a real repo_url to point to. The GitHub username used to invite the client into the task repo comes from their Stoory account, not from you — you never see or pass it. If create_task fails because it's missing or invalid, tell the user to check/set their GitHub username in their Stoory account settings, then retry create_task."

Previous value: -"The minimal set of files the specialist needs. This server runs remotely and cannot read the caller's disk, so do NOT pass local paths. Each entry is { name, url?, content_base64? } and must carry exactly one source: `url` (the server downloads it — prefer this when the user gave a link, e.g. Google Drive / Dropbox / S3 / raw GitHub) or `content_base64` (inline bytes, for small local or chat-uploaded files; raw base64 only, no `data:...;base64,` prefix). For a large binary with no URL (e.g. an image), do not inline it — tell the user to add it to the task repo manually. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. With a non-empty list the user must have github_username set — if create_task reports it missing, ask the user for their GitHub login and try again."New value: +"The minimal set of files the specialist needs. This server runs remotely and cannot read the caller's disk, so do NOT pass local paths. Each entry is { name, url?, content_base64? } and must carry exactly one source: `url` (the server downloads it — prefer this when the user gave a link, e.g. Google Drive / Dropbox / S3 / raw GitHub) or `content_base64` (inline bytes, for small local or chat-uploaded files; raw base64 only, no `data:...;base64,` prefix). For a large binary with no URL (e.g. an image), do not inline it — tell the user to add it to the task repo manually. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. A non-empty list always gets a repo, even if every entry ends up in blocked_files (unreadable or filtered) — so the 'add it manually' fallback always has a real repo_url to point to. The user must have github_username set — if create_task reports it missing, ask the user for their GitHub login and try again."

Previous value: -"The minimal set of files the specialist needs. This server runs remotely and cannot read the caller's disk, so do NOT pass local paths. Each entry is { name, url?, content_base64? } and must carry exactly one source: `url` (the server downloads it — prefer this when the user gave a link, e.g. Google Drive / Dropbox / S3 / raw GitHub) or `content_base64` (inline bytes, for small local or chat-uploaded files). For a large binary with no URL (e.g. an image), do not inline it — tell the user to add it to the task repo manually. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. With a non-empty list the user must have github_username set — if create_task reports it missing, ask the user for their GitHub login and try again."New value: +"The minimal set of files the specialist needs. This server runs remotely and cannot read the caller's disk, so do NOT pass local paths. Each entry is { name, url?, content_base64? } and must carry exactly one source: `url` (the server downloads it — prefer this when the user gave a link, e.g. Google Drive / Dropbox / S3 / raw GitHub) or `content_base64` (inline bytes, for small local or chat-uploaded files; raw base64 only, no `data:...;base64,` prefix). For a large binary with no URL (e.g. an image), do not inline it — tell the user to add it to the task repo manually. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. With a non-empty list the user must have github_username set — if create_task reports it missing, ask the user for their GitHub login and try again."

Previous value: -"Minimalny zestaw plików potrzebny specjaliście — tylko ścieżki lokalne, NIE wklejaj treści plików. Serwer sam czyta plik z dysku i wrzuca go do prywatnego repo GitHub utworzonego dla tego taska — treść nigdy nie trafia do kontekstu modelu. Nigdy całe repo, max 1 plik = ~900KB. Przy niepustej liście user musi mieć ustawiony github_username — jeśli create_task to zgłosi, dopytaj usera o jego login GitHub i spróbuj ponownie."New value: +"The minimal set of files the specialist needs. This server runs remotely and cannot read the caller's disk, so do NOT pass local paths. Each entry is { name, url?, content_base64? } and must carry exactly one source: `url` (the server downloads it — prefer this when the user gave a link, e.g. Google Drive / Dropbox / S3 / raw GitHub) or `content_base64` (inline bytes, for small local or chat-uploaded files). For a large binary with no URL (e.g. an image), do not inline it — tell the user to add it to the task repo manually. The server scans every file for secrets and pushes the rest to a private GitHub repo created for this task; contents never enter the model context. Never the whole repo, max 1 file = ~900KB. With a non-empty list the user must have github_username set — if create_task reports it missing, ask the user for their GitHub login and try again."

Added value: +{
+  "type": "string"
+}

Added value: +{
+  "type": "string"
+}

Removed value: -{
-  "type": "string"
-}

Added value: +{
+  "format": "uri",
+  "type": "string"
+}

Previous value: -[
-  "path"
-]New value: +[
+  "name"
+]

Previous value: -"Język taska — musi odpowiadać językowi, w jakim user sformułował title/description (nie ma automatycznego wykrywania, ustaw zgodnie z danymi usera)."New value: +"The task's language — must match the language the user phrased title/description in (there is no auto-detection, set it according to the user's input)."

Previous value: -"Id jednego lub kilku specjalistów do zaproszenia."New value: +"Id of one or more specialists to invite."

Previous value: -"Jak pilny jest projekt z perspektywy klienta — pytaj usera wprost, nie zgaduj."New value: +"How urgent the project is from the client's perspective — ask the user directly, do not guess."

Previous value: -"Musi być true — ustaw wyłącznie po jawnym potwierdzeniu przez usera w rozmowie"New value: +"Must be true — set it only after the user's explicit confirmation in the conversation."