feat: add model card generator workflow and configuration (#646)

* feat: add model card generator workflow and configuration

* feat: add demo command for model card generation in YAML configuration

* feat: enhance model card generator configuration with thinking budget and provider options

* feat: update model card generator template with detailed process and structure

* pin actions

* refactor(agents): define input for model-card-generator

Signed-off-by: Dorin Geman <dorin.geman@docker.com>

---------

Signed-off-by: Dorin Geman <dorin.geman@docker.com>
Co-authored-by: Dorin Geman <dorin.geman@docker.com>

Author: ilopezluna

.github/agents/model-card-generator.yaml

@@ -0,0 +1,111 @@
+models:
+  sonnet:
+    provider: anthropic
+    model: claude-sonnet-4-5
+    max_tokens: 16384
+    thinking_budget: 4096
+    provider_opts:
+        interleaved_thinking: true
+
+agents:
+  root:
+    model: sonnet
+    description: AI Model Card Generator
+    commands:
+      demo: |
+        **Repository:** aistaging/smollm2-vllm
+        **Model name:** smollm2-vllm
+        **Reference URLs to research:**
+        https://huggingface.co/HuggingFaceTB/SmolLM2-360M
+
+    instruction: |
+      You are an expert technical writer specializing in AI/ML model documentation.
+      Your task is to copy a Hugging Face repository's model card, following some guidelines and constraints.
+      Do not fetch HTML as the content is too large, use the Hugging Face API.
+
+      You'll receive these inputs:
+      ```
+      **Repository:** aistaging/${{ inputs.repository }}
+      **Model name:** ${{ inputs.repository }}
+      **Reference URLs to research:**
+      ${{ inputs.reference_urls }}
+      ```
+
+      Generate the model card and save it to: model-cards/${{ inputs.repository }}.md.
+      The model-cards directory already exists, do not check that it exists, do no attempt to create it.
+
+      At the end of the model card add:
+      ```
+      ### Generated by
+      This model card was automatically generated using [cagent-action](https://github.com/docker/cagent-action) with the [Docker Model Runner's model-card-generator](https://github.com/docker/model-runner).
+      ```
+
+      ## Process
+
+      1. **Research**: Use `curl` to fetch content from each provided reference URL to gather information about the model (architecture, parameters, benchmarks, license, capabilities, etc.)
+      2. **Generate**: Write a model card following the template below
+      3. **Save**: Write the file to the specified path using filesystem tools
+
+      ## Template
+
+      Follow this structure:
+
+      ```
+      # {Model Name}
+
+      {2-3 paragraph description: what it is, key capabilities, target use cases.}
+
+      ---
+
+      ## Characteristics
+
+      | Attribute | Value |
+      |---|---|
+      | **Provider** | {company} |
+      | **Architecture** | {arch} |
+      | **Cutoff date** | {date} |
+      | **Languages** | {languages} |
+      | **Input modalities** | {Text, Image...} |
+      | **Output modalities** | {Text, Image...} |
+      | **License** | {license} |
+      
+      ## Using this model with Docker Model Runner
+
+      ```bash
+      docker model run {model_name}
+      ```
+
+      For more information, check out the [Docker Model Runner docs](https://docs.docker.com/desktop/features/model-runner/).
+
+      ## Benchmarks
+
+      {Table with benchmark results found in reference URLs. Omit if no benchmarks found.}
+
+      ## Links
+
+      {Reference URLs as a bullet list}
+      ```
+      
+      ## Considerations
+
+      {Bullet list of limitations or recommendations. Omit if no information found.}
+
+      ## Guidelines
+
+      - **DO NOT** include Python code, scripts, or any code snippets other than `docker model` commands
+      - **DO NOT** include usage instructions for other inference engines (llama.cpp, Ollama, vLLM, TGI, HuggingFace Transformers, etc.) — mentioning them in descriptive text is fine
+      - The **only** code blocks allowed is `docker model run`
+      - Be factual: use `TBD` for information you cannot find rather than fabricating data
+      - Include all benchmark data found in the reference URLs
+      - The model card should be self-contained and ready for Docker Hub
+
+    toolsets:
+      - type: filesystem
+        tools: [read_file, write_file, list_directory]
+      - type: shell
+
+permissions:
+  allow:
+    - shell:cmd=curl *
+    - shell:cmd=cat *
+    - shell:cmd=echo *

.github/workflows/generate-model-card.yml

@@ -0,0 +1,68 @@
+name: Generate Model Card
+
+on:
+  workflow_dispatch:
+    inputs:
+      repository:
+        description: 'Docker Hub repository name in aistaging namespace (e.g., "qwen3")'
+        required: true
+        type: string
+      reference_urls:
+        description: 'Comma-separated reference URLs (HuggingFace pages, blog posts, papers, etc.)'
+        required: true
+        type: string
+
+permissions:
+  contents: write
+  pull-requests: write
+
+jobs:
+  generate:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd
+
+      - name: Generate model card
+        uses: docker/cagent-action@bd5439a1cabc3c448abf7a0d29b6f9a0d5fd4ec7
+        with:
+          agent: .github/agents/model-card-generator.yaml
+          anthropic-api-key: ${{ secrets.ANTHROPIC_API_KEY }}
+          prompt: |
+            **Repository:** aistaging/${{ inputs.repository }}
+            **Model name:** ${{ inputs.repository }}
+            **Reference URLs to research:**
+            ${{ inputs.reference_urls }}
+
+      - name: Verify model card was generated
+        run: |
+          if [ ! -f "model-cards/${{ inputs.repository }}.md" ]; then
+            echo "::error::Model card file was not generated at model-cards/${{ inputs.repository }}.md"
+            exit 1
+          fi
+          echo "✅ Model card generated successfully:"
+          head -20 "model-cards/${{ inputs.repository }}.md"
+
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@22a9089034f40e5a961c8808d113e2c98fb63676
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          commit-message: "Add model card for aistaging/${{ inputs.repository }}"
+          title: "Add model card for aistaging/${{ inputs.repository }}"
+          body: |
+            ## 🤖 Auto-generated Model Card
+
+            This PR adds a model card for **aistaging/${{ inputs.repository }}**.
+
+            ### Reference URLs used
+            ${{ inputs.reference_urls }}
+
+            ### Generated by
+            This model card was automatically generated using [cagent-action](https://github.com/docker/cagent-action) with the [Docker Model Runner's model-card-generator](https://github.com/docker/model-runner).
+
+            ---
+            Please review the generated content for accuracy before merging.
+          branch: model-card/${{ inputs.repository }}
+          base: main
+          labels: model-card,automated
+          delete-branch: true

model-cards/.gitkeep

@@ -0,0 +1 @@
+