Update to Olbrain Alchemist

Changed files

Choose a file to see its text changes. Binary or large files show metadata only.

.codex-plugin/mcp.json →

.codex-plugin/plugin.json →

.mcp.json →

skills/alchemist/SKILL.md →

skills/build/SKILL.md →

skills/status/SKILL.md →

.codex-plugin/plugin.json

− removed · + added

--- before
+++ after
@@ -1,6 +1,6 @@
 {
   "name": "alchemist",
-  "version": "0.8.6",
+  "version": "0.9.11",
   "description": "Build, test and publish Olbrain AI agents from Codex, with Olbrain's Alchemist builder.",
   "author": {
     "name": "Olbrain",
@@ -8,7 +8,7 @@
   },
   "homepage": "https://github.com/Olbrain/alchemist-plugin",
   "repository": "https://github.com/Olbrain/alchemist-plugin",
-  "license": "UNLICENSED",
+  "license": "MIT",
   "keywords": [
     "olbrain",
     "agents",
@@ -16,11 +16,11 @@
     "mcp"
   ],
   "skills": "./skills/",
-  "mcpServers": "./.mcp.json",
+  "mcpServers": "./.codex-plugin/mcp.json",
   "interface": {
     "displayName": "Olbrain Alchemist",
     "shortDescription": "Build and publish AI agents",
-    "longDescription": "Alchemist is Olbrain's agent builder. Tell Codex what the agent should do, in plain words. Alchemist's own model works out what kind of agent it is, plans it, builds it in your Olbrain organisation, tests it on real inputs and publishes it, asking you before anything goes live.\n\nIt builds chat agents, automatic jobs (started by a schedule, a webhook, an email or a new file in a watched folder), teams of agents behind one front door, and research agents. It connects website chat, WhatsApp, Shopify storefront chat, your phone number and Microsoft Teams, tools from Olbrain's connector catalog, any REST API or your own MCP server, and the agent's knowledge files.\n\nIt also shows an agent as a diagram, reviews live agents, puts a live agent on your website or behind a server route in your project, and handles your Olbrain support tickets. Every agent runs on Olbrain and stays visible in Noesis, Olbrain's console, where your team can open, review and change it.",
+    "longDescription": "Alchemist is Olbrain's agent builder. Tell Codex what the agent should do, in plain words. Codex plans and designs the agent from your words and your project, with Alchemist's step-by-step guidance; Alchemist builds it in your Olbrain organisation, tests it on real inputs, checks every step in order and publishes it, asking you before anything goes live.\n\nIt builds chat agents, automatic jobs (started by a schedule, a webhook, an email or a new file in a watched folder), teams of agents behind one front door, and research agents. It connects website chat, WhatsApp, Shopify storefront chat, your phone number and Microsoft Teams, tools from Olbrain's connector catalog, any REST API or your own MCP server, and the agent's knowledge files.\n\nIt also shows an agent as a diagram, reviews live agents, puts a live agent on your website or behind a server route in your project, and handles your Olbrain support tickets. Every agent runs on Olbrain and stays visible in Noesis, Olbrain's console, where your team can open, review and change it.",
     "developerName": "Olbrain",
     "category": "Developer Tools",
     "capabilities": [
@@ -40,8 +40,7 @@
     "brandColor": "#111111",
     "brandColorDark": "#EEEEEE",
     "composerIcon": "./.claude-plugin/icon.png",
-    "logo": "./.claude-plugin/icon.png",
-    "screenshots": []
+    "logo": "./.claude-plugin/icon.png"
   },
   "extensions": {
     "com.openai": {
@@ -63,8 +62,20 @@
             {
               "description": "Starts a build and waits for the user on Alchemist's approval cards.",
               "prompt": "Use Alchemist to build a chat agent that answers questions about our opening hours (9am to 6pm, Monday to Saturday) and hands refund requests to a person.",
-              "tools_triggered": "whoami, alchemist, alchemist_answer",
-              "expected_behavior": "Starts a build and shows Alchemist's reply. When Alchemist sends a card, such as a plan to approve, it shows the card's title, summary and question with its choices and waits for the user's answer before calling alchemist_answer. Nothing is published without a card the user approves."
+              "tools_triggered": "whoami, alchemist, alchemist_build, alchemist_answer",
+              "expected_behavior": "Starts a build by passing the user's words to alchemist, and follows the build with alchemist_build while Alchemist plans it. When a card arrives, such as a plan to approve, it shows the card's title, summary and question with its choices and waits for the user's answer before calling alchemist_answer. Nothing is published without a card the user approves."
+            },
+            {
+              "description": "A long turn is followed to the end, never sent twice.",
+              "prompt": "Use Alchemist to build a chat agent for our opening hours, and keep me posted while it works.",
+              "tools_triggered": "alchemist, alchemist_build",
+              "expected_behavior": "When alchemist returns status still_running, it says in a line what Alchemist is doing and calls alchemist_build with wait until the turn has finished, then reports the result or the card waiting. It does not send the request again or start a second build while the turn runs."
+            },
+            {
+              "description": "Removing an agent waits for the user's own yes on the remove card.",
+              "prompt": "Remove our old FAQ agent from Olbrain.",
+              "tools_triggered": "alchemist_connect, alchemist, alchemist_build",
+              "expected_behavior": "Finds the agent with alchemist_connect and starts a build on it with alchemist. When the remove card arrives, it shows the card's title, summary and consequences in full and asks its question word for word with each choice. It calls alchemist_answer only after the user answers, and the agent is not removed before the user says yes."
             },
             {
               "description": "Finds earlier builds and reports where the latest one stands.",
@@ -97,7 +108,7 @@
         "commerce": false
       },
       "publication": {
-        "release_notes": "First release for Codex and ChatGPT: build, test and publish Olbrain AI agents (chat agents, automatic jobs, teams of agents and research agents) with Alchemist, asking before anything goes live."
+        "release_notes": "Codex now plans and designs the agent itself, with step-by-step guidance from alchemist_state, and sends each step through alchemist_do; Alchemist builds, tests, checks the order of steps and publishes. Codex can take over a build that Alchemist's own model was driving. Removing an agent, forgetting a customer, renames and publishing despite failing tests are answered in the browser. Long turns return while they run, with a progress line every few seconds. Waiting builds are resumed first, builds can be renamed and rated, and a published version of an agent can be shown as a diagram. Platform gaps are reported directly instead of asking the user to raise a ticket. Errors from Alchemist now carry codes the skill acts on, so Codex recovers from a refused step on its own instead of retrying blindly. The plugin is now MIT licensed."
       }
     }
   },