@zackbart/connecta 0.28.1 → 0.29.0

package/dist/providers/cloudflare/skill.generated.js

@@ -0,0 +1,10 @@
+// Generated by scripts/generate-providers.mjs; do not edit.
+export const skill = {
+    "content": "- This is the explicit `surface: \"api\"` complement. Hosted MCP is the default; configure each under a separate connector id with independent authorization and recovery. Never fall back to another credential after an error or repeat a write to recover its response.\n- Named tools are canonical for reads in read-only programs. `cloudflare_api_get` covers other GET endpoints, including text and byte-preserving `responseType: \"base64\"` downloads. Hosted `search` discovers API contracts; hosted `execute` owns ordinary JSON mutations. Execute remains a write even for a GET-only program.\n- `cloudflare_api_mutate` retains only a demonstrated auth/header gap: Global API Key identity, or R2 bucket endpoints (including bucket creation at the collection) requiring `cf-r2-jurisdiction`, or bucket storage-class edits with `PATCH /accounts/{accountId}/r2/buckets/{bucketName}` and `cf-r2-storage-class: Standard` or `InfrequentAccess`. The vendor request helper cannot send either header. Storage-class edits do not require a jurisdiction header for ordinary buckets. Ordinary headers such as `Accept` do not qualify. Other API-token JSON mutations are refused locally and must use hosted `execute`.\n- `cloudflare_api_upload` retains byte-safe upload bodies, explicit endpoint headers and multipart field/file encoding for Worker scripts, KV values, R2 objects, Images, Stream and Pages uploads. PUT owns Worker scripts, KV values and R2 objects; POST owns Worker versions, Images, Stream collection uploads and Pages uploads. Stream metadata edits, copy-from-URL and direct-upload URL creation use hosted `execute`. Other method/path pairs are refused, regardless of claimed content type. The vendor supports raw/multipart bodies, but its helper returns non-JSON bodies as text and exposes only content type on the request. Use API uploads when those compatibility requirements matter. Raw tools take a path below `/client/v4`; absolute URLs, traversal and embedded query strings are refused. `Authorization`, `Cookie`, `Host`, `Content-Length`, `Content-Type`, and `Transfer-Encoding` are connector-owned headers and cannot be overridden.\n- Useful paths include `/accounts/{accountId}/images/v1`, `/accounts/{accountId}/stream`, `/accounts/{accountId}/d1/database`, `/accounts/{accountId}/queues`, `/accounts/{accountId}/r2/buckets/{bucketName}/cors`, and `/accounts/{accountId}/r2/metrics`. Change CORS with hosted `execute`, except for the explicit legacy-identity/header gaps above.\n- Lists use `page`/`perPage` and return `page.hasMore`; rulesets, R2 buckets/objects and KV keys instead use opaque cursors. `list_worker_scripts` is unpaginated. Use `get_zone_setting` for one setting; Cloudflare deprecated the bulk `/zones/{zoneId}/settings` read. Reads project resource identity and state; `raw: true` returns dropped vendor fields.\n- - Honor vendor rate-limit waits. The documented 1,200 calls per five minutes are cumulative across a user and its tokens.\n- Read state before replacement, rollback or purge. Prefer targeted cache purges in hosted `execute`. Mutations/uploads stay write-classified; approval belongs to the host and program access to the configured pool.\n\n\n- Hosted `execute` owns ordinary mutations. The separate `surface: \"api\"`\n  complement owns read-only-program reads, binary/header-compatible uploads\n  and legacy Global API Key or R2 jurisdiction/storage-class mutations. Configure distinct\n  connector ids and authorize each explicitly; there is no credential fallback.\n- Resolve account and zone ids independently. Defaults do not constrain token permissions. Distinguish missing or revoked credentials from missing product permissions, and honor supplied rate-limit waits.\n- The catalog contains `search` and `execute`. Search runs code against\n  Cloudflare's OpenAPI document. Execute runs code that may call any authorized\n  Cloudflare API endpoint.\n- Use `search` to find the exact method, path, and fields before writing an\n  execute program. Do not guess an endpoint from product naming.\n- Connecta classifies `execute` as a write because it can mix GET, POST,\n  PUT, PATCH, and DELETE requests inside one program. Fetch\n  `skills({ name: \"usage\" })` for guest calls and routing. The MCP schema\n  cannot establish that arbitrary code is read-only.\n- Keep returned values small. Filter and project inside the Cloudflare MCP\n  program, then reduce again inside Connecta's `execute_code` when several\n  calls must be joined.\n- OAuth and API-token permissions remain the provider-side boundary. An\n  `auth_required` failure needs authorization or a token with the required\n  Cloudflare permission.\n",
+    "instructionsHeading": "Account instructions",
+    "fragments": {
+        "guide_0": "- This is the explicit `surface: \"api\"` complement. Hosted MCP is the default; configure each under a separate connector id with independent authorization and recovery. Never fall back to another credential after an error or repeat a write to recover its response.\n- Named tools are canonical for reads in read-only programs. `cloudflare_api_get` covers other GET endpoints, including text and byte-preserving `responseType: \"base64\"` downloads. Hosted `search` discovers API contracts; hosted `execute` owns ordinary JSON mutations. Execute remains a write even for a GET-only program.\n- `cloudflare_api_mutate` retains only a demonstrated auth/header gap: Global API Key identity, or R2 bucket endpoints (including bucket creation at the collection) requiring `cf-r2-jurisdiction`, or bucket storage-class edits with `PATCH /accounts/{accountId}/r2/buckets/{bucketName}` and `cf-r2-storage-class: Standard` or `InfrequentAccess`. The vendor request helper cannot send either header. Storage-class edits do not require a jurisdiction header for ordinary buckets. Ordinary headers such as `Accept` do not qualify. Other API-token JSON mutations are refused locally and must use hosted `execute`.\n- `cloudflare_api_upload` retains byte-safe upload bodies, explicit endpoint headers and multipart field/file encoding for Worker scripts, KV values, R2 objects, Images, Stream and Pages uploads. PUT owns Worker scripts, KV values and R2 objects; POST owns Worker versions, Images, Stream collection uploads and Pages uploads. Stream metadata edits, copy-from-URL and direct-upload URL creation use hosted `execute`. Other method/path pairs are refused, regardless of claimed content type. The vendor supports raw/multipart bodies, but its helper returns non-JSON bodies as text and exposes only content type on the request. Use API uploads when those compatibility requirements matter. Raw tools take a path below `/client/v4`; absolute URLs, traversal and embedded query strings are refused. `Authorization`, `Cookie`, `Host`, `Content-Length`, `Content-Type`, and `Transfer-Encoding` are connector-owned headers and cannot be overridden.\n- Useful paths include `/accounts/{accountId}/images/v1`, `/accounts/{accountId}/stream`, `/accounts/{accountId}/d1/database`, `/accounts/{accountId}/queues`, `/accounts/{accountId}/r2/buckets/{bucketName}/cors`, and `/accounts/{accountId}/r2/metrics`. Change CORS with hosted `execute`, except for the explicit legacy-identity/header gaps above.\n- Lists use `page`/`perPage` and return `page.hasMore`; rulesets, R2 buckets/objects and KV keys instead use opaque cursors. `list_worker_scripts` is unpaginated. Use `get_zone_setting` for one setting; Cloudflare deprecated the bulk `/zones/{zoneId}/settings` read. Reads project resource identity and state; `raw: true` returns dropped vendor fields.\n- ",
+        "guide_1": "- Honor vendor rate-limit waits. The documented 1,200 calls per five minutes are cumulative across a user and its tokens.\n- Read state before replacement, rollback or purge. Prefer targeted cache purges in hosted `execute`. Mutations/uploads stay write-classified; approval belongs to the host and program access to the configured pool.\n",
+        "guide_2": "\n\n- Hosted `execute` owns ordinary mutations. The separate `surface: \"api\"`\n  complement owns read-only-program reads, binary/header-compatible uploads\n  and legacy Global API Key or R2 jurisdiction/storage-class mutations. Configure distinct\n  connector ids and authorize each explicitly; there is no credential fallback.\n- Resolve account and zone ids independently. Defaults do not constrain token permissions. Distinguish missing or revoked credentials from missing product permissions, and honor supplied rate-limit waits.\n- The catalog contains `search` and `execute`. Search runs code against\n  Cloudflare's OpenAPI document. Execute runs code that may call any authorized\n  Cloudflare API endpoint.\n- Use `search` to find the exact method, path, and fields before writing an\n  execute program. Do not guess an endpoint from product naming.\n- Connecta classifies `execute` as a write because it can mix GET, POST,\n  PUT, PATCH, and DELETE requests inside one program. Fetch\n  `skills({ name: \"usage\" })` for guest calls and routing. The MCP schema\n  cannot establish that arbitrary code is read-only.\n- Keep returned values small. Filter and project inside the Cloudflare MCP\n  program, then reduce again inside Connecta's `execute_code` when several\n  calls must be joined.\n- OAuth and API-token permissions remain the provider-side boundary. An\n  `auth_required` failure needs authorization or a token with the required\n  Cloudflare permission.\n"
+    }
+};

package/dist/providers/docs/index.d.ts

@@ -0,0 +1,54 @@
+import { type GoogleWorkspaceOptions } from "../_shared/google/workspace.js";
+export type { GoogleServiceAccount, GoogleSubjectContext, GoogleWorkspaceOptions, GoogleWorkspaceSubject, } from "../_shared/google/workspace.js";
+/**
+ * The Docs API root. Unlike Gmail there is no per-user path segment: the
+ * delegated token's subject decides which documents a request can reach.
+ */
+export declare const DOCS_API_BASE_URL = "https://docs.googleapis.com/v1";
+/**
+ * Exactly the scope this connection requests, and exactly what the Admin
+ * console's domain-wide delegation entry must list for it. `documents` reads
+ * and edits Docs the user can reach; it lists nothing and touches no other
+ * Drive file type.
+ */
+export declare const DOCS_SCOPES: readonly ["https://www.googleapis.com/auth/documents"];
+/** Options for {@link docs}: the shared Workspace delegation options. */
+export type DocsOptions = GoogleWorkspaceOptions;
+/**
+ * A maintained Google Docs connection acting as each signed-in Workspace user
+ * through a delegated service account.
+ *
+ * Setup, once per Workspace (shared with every Workspace provider):
+ *
+ * 1. In a Google Cloud project, enable the **Google Docs API**.
+ * 2. Create a service account. Grant it no IAM roles: delegation, not project
+ *    permission, is what reaches documents.
+ * 3. Create a JSON key for it. If key creation is refused, the organization
+ *    policy `iam.disableServiceAccountKeyCreation` is enforced; an org policy
+ *    administrator can override it for this project alone.
+ * 4. Copy the service account's numeric **client ID** (the key's `client_id`).
+ * 5. As a Workspace super admin: Admin console → Security → Access and data
+ *    control → API controls → Manage Domain Wide Delegation → Add new (or edit
+ *    the service account's existing entry). Paste the client ID and add
+ *    exactly this scope ({@link DOCS_SCOPES}):
+ *    `https://www.googleapis.com/auth/documents`. One entry lists every
+ *    Workspace provider's scopes, comma-separated. A new grant can take up to
+ *    24 hours to apply.
+ * 6. Configure inbound auth: a `subject` function is never asked about an
+ *    open deployment's anonymous requests, which fail `auth_required`.
+ *    Then give the deployment the key as a secret and map callers to accounts:
+ *
+ * ```ts
+ * docs("docs", {
+ *   purpose: "Staff meeting notes and sermon drafts",
+ *   serviceAccount: env.GOOGLE_SERVICE_ACCOUNT_JSON,
+ *   subject: (identity) => accounts[identity.principal?.id ?? ""],
+ * });
+ * ```
+ *
+ * Reads run in programs. `create_document`, `append_text`, and `insert_text`
+ * are additive writes the host approves through call_destructive_tool or a
+ * trusted program; `replace_all_text` and `batch_update_document` are
+ * destructive.
+ */
+export declare const docs: import("../../provider.js").ProviderFactory<GoogleWorkspaceOptions>;