@aipermission/mcp 0.2.18 → 0.2.20

package/README.md

@@ -113,10 +113,14 @@ previews and published payloads can contain secrets or customer data; use short
 reasons and prefer approval-required access until the workflow is trusted.
 
 For Kafka or Redpanda, call `get_connector_actions(target_ref)` to discover
-cluster, topic, consumer-group, lag, and bounded message-read actions. Message
-reads use explicit partition assignment, never join a consumer group, and
-never commit offsets. Payloads can contain sensitive application data, so keep
-these reads in Prompt mode until the workflow is trusted.
+cluster, topic, consumer-group, lag, bounded message-read, guarded publish, and
+single-partition offset actions. Message reads use explicit partition
+assignment, never join a consumer group, and never commit offsets. Publishing
+is a write action; consumer-group offset changes are destructive and reject
+active classic groups and modern consumer-protocol groups. Payloads can contain
+sensitive application data, and offset changes can replay or skip records, so
+keep these actions in Prompt mode unless direct execution is intentional. If a
+publish has an unknown delivery outcome, inspect before retrying.
 
 For Docker, call `get_connector_actions(target_ref)` to discover bounded
 actions such as `docker_version`, `list_containers`, `list_images`,

package/dist/resources/aipermission-operator/SKILL.md

@@ -56,6 +56,14 @@ RabbitMQ messages only when the operator explicitly asked for a write.
 For Kafka / Redpanda, discover topics before describing partitions or reading
 messages. Keep message reads bounded, prefer Prompt because payloads can be
 sensitive, and remember that read actions never join groups or commit offsets.
+Use `publish_message` only for one explicitly requested topic partition. Treat
+`set_consumer_group_offset` as destructive because it can replay or skip
+records; keep both actions on Prompt unless direct execution is intentional.
+If a publish reports an unknown delivery outcome, inspect the topic before
+retrying. Offset changes support inactive classic groups and reject modern
+consumer-protocol groups. The inactive-group check is best effort because
+Kafka cannot atomically prevent a member joining between the final check and
+commit; inspect the returned post-commit state and warning.
 
 Avoid vague reasons:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aipermission/mcp",
-  "version": "0.2.18",
+  "version": "0.2.20",
   "mcpName": "io.github.aipermission/aipermission-mcp",
   "description": "Local-first MCP bridge for the aipermission gateway.",
   "license": "AGPL-3.0-only",

package/server.json

@@ -3,12 +3,12 @@
   "name": "io.github.aipermission/aipermission-mcp",
   "title": "AIPermission",
   "description": "Local-first MCP bridge for the AIPermission gateway.",
-  "version": "0.2.18",
+  "version": "0.2.20",
   "packages": [
     {
       "registryType": "npm",
       "identifier": "@aipermission/mcp",
-      "version": "0.2.18",
+      "version": "0.2.20",
       "transport": {
         "type": "stdio"
       }