diff --git a/.github/workflows/test-pr.yml b/.github/workflows/test-pr.yml
index e4452bb..3d30d1b 100644
--- a/.github/workflows/test-pr.yml
+++ b/.github/workflows/test-pr.yml
@@ -24,7 +24,7 @@ jobs:
     
     - name: Install dependencies
       run: |
-        uv sync --dev
+        uv sync --extra test
     
     - name: Run e2e tests
       run: |
diff --git a/README.md b/README.md
index 539e6e5..66a6ac0 100644
--- a/README.md
+++ b/README.md
@@ -4,7 +4,7 @@ ContextKit is a CLI tool and MCP client, used to create spec files for AI coding
 ContextKit aims to simplify and automate the process of creating spec files for AI coding agents. Instead of creating or modifying spec files for each task in a project, ContextKit allows developers to define reusable spec templates that can be filled with context from various sources, either automatically or by providing just the necessary task-specific context variables, such as a ticket ID or Figma URL.
 
 ## How it works
-By being an MCP client, ContextKit utilizes the evolving MCP ecosystem to inject context into spec files from various sources, such as ticketing systems, databases, document storage, and design tools. It works with spec templates - reusable spec files, containing context variables, that define the common structure and requirements of tasks in a project. The context variables are either full or partial MCP resources and tools.
+By being an MCP client, ContextKit utilizes the evolving MCP ecosystem to inject context into spec files from various sources, such as ticketing systems, databases, document storage, and design tools. It works with spec templates - reusable spec files, containing context variables, that define the common structure and requirements of tasks in a project. The context variables are either fully specified or partially specified MCP resources and tools.
 
 If a context variable can be fulfilled by fetching an MCP resource, then it will be automatically fetched and injected into the spec file when generating the spec. If the context variable is a partial MCP resource, the user will be prompted to complete it before the spec file is generated. Usually the missing arguments in the MCP resource or tool are task-specific, such as a ticket ID or a Figma URL.