role is optional in h1 prompt headers

- defaults to user

Author: slimslenderslacks

docs/content/_index.md

@@ -10,5 +10,5 @@ breadcrumbs: false
 {{< cards >}}
   {{< card link="tools" title="AI Tools for Devs" subtitle="Using Docker to extend AI systems with custom tool runtimes" image="https://spec.modelcontextprotocol.io/images/dark.svg" tag="Incubating">}}
   {{< card link="speculative" title="Speculative Execution" image="img/speculative.png" tag="Incubating">}}
-  {{< card link="projects/vscode.md" title="Docker VSCode Extension" image="img/vscode.png" tag="Engineering">}}
+  {{< card link="projects/vscode" title="Docker VSCode Extension" image="img/vscode.png" tag="Engineering">}}
 {{< /cards >}}

docs/content/projects/vscode.md

@@ -0,0 +1,6 @@
+
+# Background
+
+We worked on a VSCode extension to add a new Docker Language Service. The language sevice is IDE agnostic and can be attached to IDEs like Intellij, and neovim.
+
+The [current extension releases](https://github.com/docker/docker-vscode-extension/releases) are available internally to Docker employees.

docs/content/tools/_index.md

@@ -1,4 +1,10 @@
 ---
-title: Tools For Devs (Model Context Protocol)
+title: Tools For Devs
 weight: 1
 ---
+
+## AI Tools for Dev
+
+
+
+## Model Context Protocol

docs/content/tools/docs/_index.md

@@ -1,5 +1,5 @@
 ---
-title: Documentation
+title: Tools Project Documentation
 ---
 
 

docs/content/tools/docs/authoring-prompts.md

@@ -1,3 +1,8 @@
+---
+title: Authoring Prompts
+weight: 2
+---
+
 # Prompt files
 
 A prompt is markdown content with a preamble describing tools available to the agent when executing this prompt.
@@ -107,3 +112,96 @@ Run the curl command, in silent mode, to fetch gists for user slimslenderslacks
 Set streaming to false if you're using Ollama for tool calling. Ollama does not currently stream tool calls.
 {{< /callout >}}
 
+## Prompt Templates
+
+It is common for prompts to contain parameters that are either extracted from a user interaction 
+or, in the case of RAG, are populated by some sort of retrieval process. Markdown prompts can also
+contain template parameters.
+
+For example, the above curl example could be re-written as a template with a ``{{ user }}`` parameter.
+
+```markdown
+---
+tools:
+  - name: curl
+url:  http://localhost/v1/chat/completions
+stream: false
+model: llama3.1
+---
+
+# prompt
+
+Run the curl command, in silent mode, to fetch gists for user {{ user }} from GitHub.
+```
+
+### Binding values during testing
+
+When running in VSCode, you can set values of the parameters in the markdown preamble.
+
+```markdown
+---
+parameter-values:
+  user: slimslenderslacks
+---
+```
+
+### Extractors
+
+Extractors are container functions that can be used to extract values when the prompt has been deployed
+to a server. These extractors are also used to populate default values for a prompt when it is used from
+an MCP client.
+
+Extractor definitions also follow the pattern of compose services. They are just docker images but with 
+the additional requirement that they should write `application/json` to stdout. This json will be used to
+populate the context for binding parameters in the prompt template.
+
+```markdown
+---
+extractors:
+  - name: linguist
+    image: vonwig/go-linguist:latest
+    command:
+      - -json
+---
+```
+
+We can create lists if the extractor json output has array types.  For example,
+if we run the linguist tool to extract language from a project, our prompt can list
+them using the following template.  You need to be familar with the json format output
+by linguist (eg that it creates lists of maps with a `language` key).
+
+```markdown
+---
+extractors:
+  - name: linguist 
+---
+
+# prompt
+
+{{#linguist}}
+
+This project contains {{language}} code.
+
+{{/linguist}}
+
+```
+
+### Template Engine
+
+We support two templating engines today.
+
+* [mustache](https://mustache.github.io/mustache.5.html) is the default
+* [django](https://docs.djangoproject.com/en/5.1/topics/templates/)
+
+If you want to use django, then add the following field in the markdown preamble.
+
+```markdown
+---
+prompt-format: "django"
+---
+```
+
+### MCP arguments
+
+
+

docs/content/tools/docs/claude-desktop.md

@@ -1,5 +1,6 @@
 ---
 title: Using Claude Desktop
+weight: 3
 ---
 
 Enable mcp_run in your claude_desktop_config.json file using the following snippet.  See the [quickstart for Claude Desktop Users](https://modelcontextprotocol.io/quickstart/user) for more details.

docs/content/tools/docs/quickstart.md

@@ -0,0 +1,6 @@
+---
+title: Quick Start
+weight: 1
+---
+
+

docs/content/tools/mcp.md

@@ -0,0 +1,3 @@
+---
+title: Model Context Protocol
+---

prompts/examples/curl.md

@@ -4,7 +4,7 @@ tools:
   - name: curl
 ---
 
-# prompt user
+# prompt
 
 Run the curl command, in silent mode, to fetch gists for user slimslenderslacks from GitHub.
 

src/markdown.clj

@@ -31,11 +31,12 @@
          (update 0 (fn [s] (.substring ^String s (dec c1))))
          (update (dec (count lines)) (fn [s] (.substring ^String s 0 (- c2 c1))))))))
 
-(def prompt-pattern #"(?i)\s*prompt\s+(\w+)\s?")
+(def prompt-pattern-with-role-capture #"(?i)\s*prompt\s+(\w+)\s?")
+(def prompt-pattern #"(?i)\s*prompt\s?(\w+)?\s?")
 
 (defn extract-role [s]
   (second
-   (re-find prompt-pattern s)))
+   (re-find prompt-pattern-with-role-capture s)))
 
 ;; headings that include the word Prompt
 (defn prompt-section? [content node]
@@ -51,10 +52,18 @@
 ;; extract Role from Prompt ....
 (defn node-content [content node]
   {:role
-   (-> node (nth 2) (nth 3) (nth 1) (from-range content) (extract-role))
+   (or (-> node (nth 2) (nth 3) (nth 1) (from-range content) (extract-role)) "user")
    :content
    (remove-first-line (from-range (nth node 1) content))})
 
+(comment
+  (extract-role "prompt user")
+  (extract-role "prompt")
+  (extract-role "prompt user and more")
+  (re-matches prompt-pattern "prompt")
+  (re-matches prompt-pattern "prompt user")
+  (re-matches prompt-pattern "prompt user"))
+
 (defn extract-prompts [content ast]
   (->>
    (iterate zip/next (zip/seq-zip ast))