@kaminari-ad/mcp 0.9.1 → 0.11.0

package/CHANGELOG.md

@@ -7,6 +7,105 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.11.0] - 2026-07-28
+
+### Added
+
+- **Combo-rule match scope.** `create_custom_rule`, `update_custom_rule`, and
+  `test_custom_rule` now document the rule-level `config.match_scope` key used
+  by `rule_type='combo'`: `"scan"` (the default, and the behaviour when the key
+  is absent) counts conditions across the union of all tags on the scan, while
+  `"url"` requires every condition to be satisfied by tags on the same link and
+  assigns the output tag to that link. The description warns about the traps a
+  per-link rule falls into: a combo built only from link-less tags (AI verdicts,
+  crawler behaviour) has no link to attach to and never matches, AI verdicts and
+  per-link detections are not evaluated together today, and a per-link rule needs
+  at least one positive condition. Any other `match_scope` value is rejected
+  before the request is sent, so a misspelled scope surfaces as a tool-input
+  error instead of a round trip; the API validates `match_scope` on create,
+  update, and the rule-test preview as well, so this is early feedback rather
+  than the durable guarantee. The rest of a combo config is still validated only
+  for admin-authored system rules, which is why the per-link traps are documented
+  rather than enforced. `get_custom_rule` and `list_custom_rules` now flag that
+  the returned `config` may carry `match_scope` and that it must be resent
+  verbatim, since `update_custom_rule` replaces `config` wholesale.
+- **`referrer` on scan and campaign creation.** `create_scan`,
+  `create_bulk_scans`, `create_campaign`, and `update_campaign` accept an
+  optional http(s) page URL the check is performed from. With `ad_tag` or
+  `vast_tag` it is the publisher page the tag is embedded in: the browser
+  commits the page document on that URL without fetching the publisher, so the
+  creative renders as if embedded there. With `url` or `ad_discovery` it is
+  where the visitor came from, sent as the `Referer` of the page request.
+  `update_campaign` also accepts `referrer: null` to clear a stored one.
+  Scan and campaign responses surface `referrer` too, so an agent can confirm
+  what a scan actually ran with.
+
+  **Release ordering — this depends on an unreleased API.** The snapshots were
+  generated from api `feat/scan-referrer`
+  ([api!341](https://gitlab.sdev.pw/adverif/api/-/merge_requests/341)). API
+  request DTOs are `extra="ignore"`, so an API without the field drops
+  `referrer` silently: no 422, the scan runs with no referrer, and the response
+  carries no `referrer` key — the agent gets a green result for a check it
+  believes ran from a publisher page. **Do not cut the npm tag or the
+  `kaminariad-mcp` image until api!341 is merged and deployed to
+  `kaminariadprod1`.** Nothing automated enforces this.
+
+### Changed
+
+- Regenerated the OpenAPI type + zod-schema snapshots so the `/api/v1` surface
+  they describe carries `referrer`. That field is their only delta versus the
+  0.10.0 snapshots.
+
+## [0.10.0] - 2026-07-28
+
+### Added
+
+- **Repeats and retries.** `create_scan`, `create_bulk_scans`,
+  `create_campaign`, and `update_campaign` accept `repeat_count` (1-20),
+  `repeat_mode` (`isolated` | `shared`), and `retry_max_attempts` (0-5).
+  `repeat_count` multiplies the scans created — and billed — per
+  `url x country x device` combination; `shared` runs a combination's repeats
+  in one browser behind one IP with cookies carried over (rejected with 422
+  together with ad discovery); `retry_max_attempts` re-crawls the same scan
+  after a transient failure without billing twice.
+- **Repeat / retry fields on the read side.** Scan detail surfaces
+  `repeat_index`, `repeat_total`, `repeat_session_id`, `repeat_scan_ids`,
+  `retry_attempt`, and `retry_max_attempts`; the scan-list brief carries all of
+  those except `repeat_scan_ids`; campaign responses echo `repeat_count`,
+  `repeat_mode`, and `retry_max_attempts`.
+
+### Fixed
+
+- **`list_balance_history` can filter on `card_top_up`.** The API has emitted
+  that transaction type for a while — an unfiltered call already returned the
+  rows — but the tool's own enum stopped at `crypto_top_up`, so
+  `type: ["card_top_up"]` was rejected locally and never reached the API. The
+  filter's length cap was one short for the same reason.
+  `BalanceTransactionType` is now aliased off the generated schema instead of
+  hand-mirrored, so the next regen turns a new value into a compile error.
+
+### Changed
+
+- Regenerated the OpenAPI type + zod-schema snapshots against `/api/v1`
+  v1.27.0. Besides the repeat / retry fields this picks up the
+  `creative-html` / `creative-video` / `vast-xml` artifact endpoints,
+  `video.click_through`, `creative_kind` as an enum, `tag_visibility` on custom
+  rules, the `card_top_up` balance-transaction type, and the `tag_match` scan
+  filter.
+- `creative_kind` is surfaced as an open string rather than the regenerated
+  `banner | video` enum, following the same forward-compat policy as
+  `block_reason`. A creative kind added on the API side now degrades that one
+  field instead of failing the entire `get_scan` / `create_scan` /
+  `create_bulk_scans` response until an MCP release ships.
+- `get_scan_screenshot` and `get_scan_landing_screenshot` warn that a resized
+  capture is top-cropped past 2.5x its width, and to omit `width` when the
+  whole page matters. `get_scan_creative_screenshot` never crops and is
+  unchanged.
+- `get_campaign` and `list_campaigns` describe the repeat / retry settings they
+  echo, `list_scan_children` explains why those fields are always neutral on a
+  discovered-ad child, and `get_scan` documents the creative's `click_through`
+  destination.
+
 ## [0.8.0] - 2026-07-10
 
 ### Added

package/dist/bin.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { NAME, VERSION, err, ok } from './chunk-JEGBPX3R.js';
+import { NAME, VERSION, err, ok } from './chunk-XSI6SZTX.js';
 import process from 'process';
 import { z } from 'zod';
 
@@ -160,10 +160,10 @@ async function main() {
   }
   const config = configResult.value;
   if (config.transport === "stdio") {
-    const { bootstrapStdio } = await import('./stdio-bootstrap-YQFKQP2G.js');
+    const { bootstrapStdio } = await import('./stdio-bootstrap-QK4ENRPP.js');
     return bootstrapStdio(config);
   }
-  const { bootstrapHttp } = await import('./http-bootstrap-JS2O6I22.js');
+  const { bootstrapHttp } = await import('./http-bootstrap-XJNTLL2C.js');
   return bootstrapHttp(config);
 }
 main().then(