Automate browser based workflows with AI
Python aiapiautomationbrowserbrowser-automationcomputergptllmplaywrightpowerautomatepuppeteerpythonrpaseleniumvisionworkflow
add LLM-safe context to conditional NL branch evaluation (#4368)
Celal Zamanoglud643b7a81c9b
diff --git a/skyvern/forge/prompts/skyvern/conditional-prompt-branch-evaluation.j2 b/skyvern/forge/prompts/skyvern/conditional-prompt-branch-evaluation.j2index cb3923843..1682daee6 100644--- a/skyvern/forge/prompts/skyvern/conditional-prompt-branch-evaluation.j2+++ b/skyvern/forge/prompts/skyvern/conditional-prompt-branch-evaluation.j2@@ -5,6 +5,9 @@ Criteria (order matters; align outputs to these indices):- {{ criterion.index }}: {{ criterion.expression }}{% endfor %}+Context (use this data to evaluate each criterion; if a value is absent, treat it as missing/Falsey):+{{ context_snapshot }}+Respond with JSON exactly in this shape:{"branch_results": [true | false per criterion, in order]diff --git a/skyvern/forge/sdk/workflow/models/block.py b/skyvern/forge/sdk/workflow/models/block.pyindex 1ffb587b1..f383b985c 100644--- a/skyvern/forge/sdk/workflow/models/block.py+++ b/skyvern/forge/sdk/workflow/models/block.py@@ -4318,6 +4318,43 @@ class BranchEvaluationContext:self.block_label = block_labelself.template_renderer = template_renderer+ def build_llm_safe_context_snapshot(self) -> dict[str, Any]:+ """+ Build a non-secret context blob for LLM-facing branch evaluation.++ Secrets are stripped/masked; only params/outputs/environment and cached+ block metadata are included so the LLM can ground purely natural language+ expressions without requiring inline templating.+ """+ if self.workflow_run_context is None:+ return {}++ ctx = self.workflow_run_context++ # Start from the recorded values (params, outputs, env, block outputs)+ snapshot: dict[str, Any] = ctx.values.copy()++ # Add block metadata (e.g., loop indices/current_item) without mutating originals+ snapshot["blocks_metadata"] = ctx.blocks_metadata.copy()++ # Ensure the common namespaces exist+ snapshot.setdefault("params", snapshot.get("params", {}))+ snapshot.setdefault("outputs", snapshot.get("outputs", {}))+ snapshot.setdefault("environment", snapshot.get("environment", {}))+ snapshot.setdefault("env", snapshot.get("environment", {}))+ snapshot.setdefault("llm", snapshot.get("llm", {}))++ # Standard workflow identifiers for additional context+ snapshot.setdefault("workflow_title", ctx.workflow_title)+ snapshot.setdefault("workflow_id", ctx.workflow_id)+ snapshot.setdefault("workflow_permanent_id", ctx.workflow_permanent_id)+ snapshot.setdefault("workflow_run_id", ctx.workflow_run_id)++ # Mask any real secret values that may have leaked into values+ snapshot = ctx.mask_secrets_in_data(snapshot)++ return snapshot+def build_template_data(self) -> dict[str, Any]:"""Build Jinja template data mirroring block parameter rendering context."""if self.workflow_run_context is None:@@ -4560,6 +4597,8 @@ class ConditionalBlock(Block):raise ValueError("organization_id is required to evaluate natural language branches")workflow_run_context = evaluation_context.workflow_run_context+ context_snapshot = evaluation_context.build_llm_safe_context_snapshot()+ context_snapshot_json = json.dumps(context_snapshot, default=str)rendered_branch_criteria: list[dict[str, Any]] = []for idx, branch in enumerate(branches):@@ -4578,6 +4617,7 @@ class ConditionalBlock(Block):extraction_goal = prompt_engine.load_prompt("conditional-prompt-branch-evaluation",branch_criteria=branch_criteria_payload,+ context_snapshot=context_snapshot_json,)data_schema = {