Add structured task playground sandbox and resilient model fallback for task runs

.github/workflows/hats-task.yml

@@ -61,6 +61,26 @@ on:
         required: false
         type: string
         default: ""
+      category:
+        description: "Playground category (default: inferred from task type)"
+        required: false
+        type: string
+        default: ""
+      genre:
+        description: "Playground genre/type bucket"
+        required: false
+        type: string
+        default: ""
+      project:
+        description: "Playground project slug"
+        required: false
+        type: string
+        default: ""
+      workspace_root:
+        description: "Sandbox workspace root on the runner"
+        required: false
+        type: string
+        default: "/tmp/hats-playground"
 
 permissions:
   contents: read
@@ -90,17 +110,31 @@ jobs:
           if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
             # workflow_dispatch inputs are controlled by the UI schema — safe
             echo "task=${{ inputs.task }}" >> "$GITHUB_OUTPUT"
-            echo "hats=${{ inputs.hats }}" >> "$GITHUB_OUTPUT"
-            echo "callback_repo=${{ inputs.callback_repo }}" >> "$GITHUB_OUTPUT"
-            echo "callback_pr=${{ inputs.callback_pr }}" >> "$GITHUB_OUTPUT"
-            echo "callback_issue=${{ inputs.callback_issue }}" >> "$GITHUB_OUTPUT"
-            # Prompt may contain special chars — pass via env to avoid code injection
-            INPUT_PROMPT="${{ inputs.prompt }}" python3 -c "
+            # String inputs may contain newlines — sanitize before writing outputs
+            INPUT_PROMPT="${{ inputs.prompt }}" \
+            INPUT_HATS="${{ inputs.hats }}" \
+            INPUT_CALLBACK_REPO="${{ inputs.callback_repo }}" \
+            INPUT_CALLBACK_PR="${{ inputs.callback_pr }}" \
+            INPUT_CALLBACK_ISSUE="${{ inputs.callback_issue }}" \
+            INPUT_CATEGORY="${{ inputs.category }}" \
+            INPUT_GENRE="${{ inputs.genre }}" \
+            INPUT_PROJECT="${{ inputs.project }}" \
+            INPUT_WORKSPACE_ROOT="${{ inputs.workspace_root }}" \
+            python3 -c "
           import os, re
-          val = os.environ.get('INPUT_PROMPT', '')
-          sanitized = re.sub(r'[\r\n]', ' ', val).strip()
+          def sanitize(value):
+              text = str(value or '')
+              return re.sub(r'[\r\n]', ' ', text).strip()
           with open(os.environ['GITHUB_OUTPUT'], 'a') as f:
-              f.write(f'prompt={sanitized}\n')
+              f.write(f\"prompt={sanitize(os.environ.get('INPUT_PROMPT', ''))}\n\")
+              f.write(f\"hats={sanitize(os.environ.get('INPUT_HATS', ''))}\n\")
+              f.write(f\"callback_repo={sanitize(os.environ.get('INPUT_CALLBACK_REPO', ''))}\n\")
+              f.write(f\"callback_pr={sanitize(os.environ.get('INPUT_CALLBACK_PR', ''))}\n\")
+              f.write(f\"callback_issue={sanitize(os.environ.get('INPUT_CALLBACK_ISSUE', ''))}\n\")
+              f.write(f\"category={sanitize(os.environ.get('INPUT_CATEGORY', ''))}\n\")
+              f.write(f\"genre={sanitize(os.environ.get('INPUT_GENRE', ''))}\n\")
+              f.write(f\"project={sanitize(os.environ.get('INPUT_PROJECT', ''))}\n\")
+              f.write(f\"workspace_root={sanitize(os.environ.get('INPUT_WORKSPACE_ROOT', ''))}\n\")
           "
           else
             # repository_dispatch — extract from client_payload with sanitization
@@ -126,6 +160,10 @@ jobs:
               f.write(f\"callback_repo={sanitize(payload.get('callback_repo', ''))}\n\")
               f.write(f\"callback_pr={sanitize(payload.get('callback_pr', ''))}\n\")
               f.write(f\"callback_issue={sanitize(payload.get('callback_issue', ''))}\n\")
+              f.write(f\"category={sanitize(payload.get('category', ''))}\n\")
+              f.write(f\"genre={sanitize(payload.get('genre', ''))}\n\")
+              f.write(f\"project={sanitize(payload.get('project', ''))}\n\")
+              f.write(f\"workspace_root={sanitize(payload.get('workspace_root', '/tmp/hats-playground'))}\n\")
               f.write(f\"context={sanitize(payload.get('context', ''))}\n\")
           "
           fi
@@ -175,14 +213,29 @@ jobs:
             --task "${{ steps.params.outputs.task }}"
             --prompt "${{ steps.params.outputs.prompt }}"
             --config scripts/hat_configs.yml
-            --output /tmp/hats-task-output
             --json-file /tmp/hats-task-result.json
+            --workspace-root "${{ steps.params.outputs.workspace_root }}"
+            --source-repo "${{ steps.params.outputs.callback_repo }}"
+            --source-pr "${{ steps.params.outputs.callback_pr }}"
+            --source-issue "${{ steps.params.outputs.callback_issue }}"
           )
 
           if [ -n "${{ steps.params.outputs.hats }}" ]; then
             ARGS+=(--hats "${{ steps.params.outputs.hats }}")
           fi
 
+          if [ -n "${{ steps.params.outputs.category }}" ]; then
+            ARGS+=(--category "${{ steps.params.outputs.category }}")
+          fi
+
+          if [ -n "${{ steps.params.outputs.genre }}" ]; then
+            ARGS+=(--genre "${{ steps.params.outputs.genre }}")
+          fi
+
+          if [ -n "${{ steps.params.outputs.project }}" ]; then
+            ARGS+=(--project "${{ steps.params.outputs.project }}")
+          fi
+
           if [ -d "/tmp/hats-context" ] && [ "$(ls -A /tmp/hats-context 2>/dev/null)" ]; then
             ARGS+=(--context-dir /tmp/hats-context)
           fi
@@ -211,15 +264,16 @@ jobs:
           CALLBACK_PR="${{ steps.params.outputs.callback_pr }}"
           CALLBACK_ISSUE="${{ steps.params.outputs.callback_issue }}"
           TARGET="${CALLBACK_PR:-$CALLBACK_ISSUE}"
+          OUTPUT_DIR="${{ steps.run.outputs.output_dir }}"
 
           if [ -z "$GH_TOKEN" ] || [ -z "$TARGET" ]; then
             echo "⚠️ Cannot post results — missing token or target"
             exit 0
           fi
 
           # Build comment from summary
-          if [ -f /tmp/hats-task-output/HATS_TASK_SUMMARY.md ]; then
-            REPORT_BODY=$(cat /tmp/hats-task-output/HATS_TASK_SUMMARY.md)
+          if [ -n "$OUTPUT_DIR" ] && [ -f "$OUTPUT_DIR/HATS_TASK_SUMMARY.md" ]; then
+            REPORT_BODY=$(cat "$OUTPUT_DIR/HATS_TASK_SUMMARY.md")
           else
             REPORT_BODY="🎩 Hats Task completed. Files generated: ${{ steps.run.outputs.files_generated }}"
           fi
@@ -241,7 +295,15 @@ jobs:
         uses: actions/upload-artifact@v4
         with:
           name: hats-task-output
-          path: /tmp/hats-task-output/
+          path: ${{ steps.run.outputs.output_dir }}/
+          retention-days: 30
+
+      - name: Upload playground workspace
+        if: always() && steps.run.outputs.workspace_root != ''
+        uses: actions/upload-artifact@v4
+        with:
+          name: hats-playground
+          path: ${{ steps.run.outputs.workspace_root }}/
           retention-days: 30
 
       - name: Upload JSON result

FORK_SETUP.md

@@ -179,15 +179,15 @@ export HAT_STACK_REPO="YOUR_USERNAME/hat_stack"
 ```bash
 # Generate a new code module — results posted as a PR comment
 hat task generate_code "Build a FastAPI auth module with JWT and refresh tokens" \
-  --repo myorg/myapp --pr 42
+  --repo myorg/myapp --pr 42 --category code --genre api --project auth-service
 
 # Write documentation for an endpoint
 hat task generate_docs "Write API documentation for the /users endpoints" \
   --repo myorg/myapp --issue 10
 
 # Plan a migration
 hat task plan "Plan a migration from REST to GraphQL for the orders service" \
-  --repo myorg/myapp
+  --repo myorg/myapp --category plans --genre migration --project orders-service
 
 # Generate tests for a module
 hat task test "Write unit tests for auth.py covering edge cases and error paths" \
@@ -210,7 +210,26 @@ hat status
 3. The task runner selects the right hats and models for the job
 4. Primary hat generates the deliverable, supporting hats review/enhance it
 5. Gold Hat does final QA
-6. Results are posted back to your project's PR/issue as a comment
+6. Results are written into a sandboxed playground tree on the runner using `category/genre/project/run-id`
+7. Results are posted back to your project's PR/issue as a comment and uploaded as artifacts
+
+**Playground layout:**
+
+```text
+/tmp/hats-playground/
+└── <category>/
+    └── <genre>/
+        └── <project>/
+            └── <run-id>/
+                ├── generated files...
+                ├── HATS_TASK_SUMMARY.md
+                ├── hats_task_result.json
+                └── PLAYGROUND_MANIFEST.json
+```
+
+If you do not pass `--category`, `--genre`, or `--project`, Hat Stack infers sensible defaults and creates the folders automatically.
+
+If the first Ollama model fails during task mode, Hat Stack automatically retries comparable configured fallback models before giving up.
 
 **For Copilot in VS Code:** Your Copilot agent can shell out to `hat task ...` commands. The `gh` CLI handles auth, and hat_stack handles execution. Your Copilot agent gives the instruction, hat_stack's model pool does the heavy lifting, results come back to the PR.
 

README.md

@@ -190,7 +190,8 @@ cp scripts/hat /usr/local/bin/hat   # or add scripts/ to PATH
 export HAT_STACK_REPO="YOUR_USERNAME/hat_stack"
 
 # Generate code
-hat task generate_code "Build a FastAPI auth module with JWT" --repo myorg/app --pr 42
+hat task generate_code "Build a FastAPI auth module with JWT" \
+  --repo myorg/app --pr 42 --category code --genre api --project auth-service
 
 # Write documentation
 hat task generate_docs "Write API docs for /users endpoints" --repo myorg/app --issue 10
@@ -208,6 +209,16 @@ hat task analyze "Security audit of payment processing" --repo myorg/payments
 git diff main | hat review - --repo myorg/app --pr 123
 ```
 
+Task runs now support a structured playground sandbox on the runner:
+
+- Workflow default workspace root: `/tmp/hats-playground`
+- Layout: `<workspace>/<category>/<genre>/<project>/<run-id>/`
+- Contents: generated files, `HATS_TASK_SUMMARY.md`, `hats_task_result.json`, `PLAYGROUND_MANIFEST.json`
+- Persistence: both the run output and the full playground tree are uploaded as workflow artifacts
+- Resilience: if the first Ollama model fails, task mode retries comparable configured fallback models automatically
+
+If no workspace root is provided outside the workflow, task mode falls back to a unique temporary output directory under `/tmp`.
+
 Or dispatch directly via `gh` CLI (what your Copilot agent would call):
 
 ```bash
@@ -275,4 +286,3 @@ hat_stack/
 ## License
 
 MIT — See [LICENSE](LICENSE).
-

scripts/hat

@@ -9,7 +9,8 @@
 #
 # Usage:
 #   hat review <diff_file> [--repo owner/repo] [--pr 42]
-#   hat task <task_type> "<prompt>" [--repo owner/repo] [--pr 42] [--hats black,green]
+#   hat task <task_type> "<prompt>" [--repo owner/repo] [--pr 42] [--hats black,green] \
+#     [--category code] [--genre api] [--project auth-service]
 #   hat status [run_id]
 #   hat list-tasks
 #
@@ -19,7 +20,7 @@
 #
 #   # Generate a new module
 #   hat task generate_code "Build a FastAPI auth module with JWT and refresh tokens" \
-#     --repo myorg/myapp --pr 123
+#     --repo myorg/myapp --pr 123 --category code --genre api --project auth-service
 #
 #   # Write docs for an existing module
 #   hat task generate_docs "Write API documentation for the /users endpoints" \
@@ -174,6 +175,7 @@ cmd_task() {
   local task_type="$1"; shift
   local prompt="$1"; shift
   local callback_repo="" callback_pr="" callback_issue="" hats="" context=""
+  local category="" genre="" project=""
 
   while [[ $# -gt 0 ]]; do
     case "$1" in
@@ -182,6 +184,9 @@ cmd_task() {
       --issue)   callback_issue="$2"; shift 2 ;;
       --hats)    hats="$2"; shift 2 ;;
       --context) context="$2"; shift 2 ;;
+      --category) category="$2"; shift 2 ;;
+      --genre)    genre="$2"; shift 2 ;;
+      --project)  project="$2"; shift 2 ;;
       *) die "Unknown option: $1" ;;
     esac
   done
@@ -199,6 +204,7 @@ cmd_task() {
   payload=$(HAT_TASK="$task_type" HAT_PROMPT="$prompt" \
     HAT_CB_REPO="$callback_repo" HAT_CB_PR="$callback_pr" \
     HAT_CB_ISSUE="$callback_issue" HAT_HATS="$hats" HAT_CTX="$context" \
+    HAT_CATEGORY="$category" HAT_GENRE="$genre" HAT_PROJECT="$project" \
     python3 -c "
 import json, os
 payload = {
@@ -209,6 +215,9 @@ payload = {
     'callback_issue': os.environ.get('HAT_CB_ISSUE', ''),
     'hats': os.environ.get('HAT_HATS', ''),
     'context': os.environ.get('HAT_CTX', ''),
+    'category': os.environ.get('HAT_CATEGORY', ''),
+    'genre': os.environ.get('HAT_GENRE', ''),
+    'project': os.environ.get('HAT_PROJECT', ''),
 }
 payload = {k: v for k, v in payload.items() if v}
 print(json.dumps(payload))
@@ -250,7 +259,7 @@ cmd_list_tasks() {
   echo ""
   echo -e "${BOLD}Usage:${NC}"
   echo '  hat task generate_code "Build a user auth module" --repo myorg/myapp --pr 42'
-  echo '  hat task plan "Plan migration to microservices" --repo myorg/monolith'
+  echo '  hat task plan "Plan migration to microservices" --repo myorg/monolith --category plans --genre migration --project orders'
 }
 
 cmd_help() {
@@ -269,10 +278,13 @@ cmd_help() {
   echo "  --issue <number>       Issue to post results to"
   echo "  --hats <hat1,hat2>     Specific hats to use"
   echo "  --context <text>       Additional context"
+  echo "  --category <name>      Playground category label"
+  echo "  --genre <name>         Playground genre/type label"
+  echo "  --project <name>       Playground project label"
   echo ""
   echo -e "${BOLD}Examples:${NC}"
   echo '  git diff main | hat review - --repo myorg/app --pr 42'
-  echo '  hat task generate_code "Build JWT auth module" --repo myorg/app --pr 42'
+  echo '  hat task generate_code "Build JWT auth module" --repo myorg/app --pr 42 --category code --genre api --project auth'
   echo '  hat task generate_docs "Write API docs for /users" --repo myorg/app --issue 10'
   echo '  hat task plan "Plan GraphQL migration" --repo myorg/app'
   echo '  hat task analyze "Security audit of payments" --repo myorg/payments'