feat: add pipeline engine with approval flow and file triggers

Sequential step executor (script, claude_prompt, approval, api_call,
template, skyvern placeholder), reaction-based approvals, file upload
trigger matching, portal API state sync.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

pipelines/steps/claude_prompt.py

@@ -0,0 +1,32 @@
+"""Claude prompt step — call LLM via LiteLLM proxy."""
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+async def execute_claude_prompt(
+    config: dict,
+    llm=None,
+    default_model: str = "claude-haiku",
+    escalation_model: str = "claude-sonnet",
+    **_kwargs,
+) -> str:
+    """Send a prompt to Claude and return the response."""
+    if not llm:
+        raise RuntimeError("LLM client not configured")
+
+    prompt = config.get("prompt", "")
+    if not prompt:
+        raise ValueError("claude_prompt step requires 'prompt' field")
+
+    model_name = config.get("model", "default")
+    model = escalation_model if model_name == "escalation" else default_model
+
+    response = await llm.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt}],
+        max_tokens=4096,
+    )
+
+    return response.choices[0].message.content or ""