Create pack_builder.py

pipeline/pack_builder.py

@@ -0,0 +1,70 @@
+from pathlib import Path
+import json
+import re
+from .io_utils import save_json
+# Replace with your real LLM runner
+from .run_two_phase import call_model
+
+def slugify(name: str) -> str:
+    s = re.sub(r"[^a-zA-Z0-9]+", "_", name.strip()).strip("_").lower()
+    return s or "scenario"
+
+def build_pack_from_scenario(root_dir: Path, scenario_text: str, pack_name_hint: str = "") -> Path:
+    """
+    1) Calls extractor prompts to create inputs.json from free text.
+    2) Writes default constraints.json and schema.json.
+    3) Returns the created pack path.
+    """
+    prompts_dir = root_dir / "prompts"
+    extractor_sys = (prompts_dir / "extractor_system.txt").read_text(encoding="utf-8")
+    extractor_user_tmpl = (prompts_dir / "extractor_user_template.txt").read_text(encoding="utf-8")
+    user_prompt = extractor_user_tmpl.replace("{SCENARIO_TEXT}", scenario_text)
+
+    raw = call_model(extractor_sys, user_prompt)
+    try:
+        extracted = json.loads(raw)
+    except Exception as e:
+        raise ValueError(f"Extractor produced non-JSON output.\nRaw:\n{raw}") from e
+
+    # Decide pack name
+    hint = pack_name_hint or extracted.get("context") or "scenario"
+    pack_slug = slugify(hint)[:48]
+    pack_dir = root_dir / "packs" / pack_slug
+    pack_dir.mkdir(parents=True, exist_ok=True)
+
+    # inputs.json
+    save_json(pack_dir / "inputs.json", extracted)
+
+    # constraints.json (default, can be edited later)
+    default_constraints = {
+        "analytics_first": True,
+        "require_longitudinal": True,
+        "indicator_priority": ["obesity_pct", "metabolic_syndrome_pct", "membership_2021"],
+        "capacity_formula": "teams * clients_per_team_per_day * working_days",
+        "cost_formula": "startup_per_client + ongoing_per_client",
+        "prioritization_rule": "Sort by membership_2021 and comorbidity prevalence; tie-breaker: infrastructure"
+    }
+    save_json(pack_dir / "constraints.json", default_constraints)
+
+    # schema.json (points to global analysis schema)
+    save_json(pack_dir / "schema.json", {
+        "archetypes": ["burden_prioritization", "capacity", "cost", "outcomes", "optimization"],
+        "output_schema": "schemas/analysis_output.schema.json"
+    })
+
+    # placeholders to support pipeline (optional for new packs)
+    save_json(pack_dir / "rubric.json", {
+        "set_equals": [],
+        "must_contain": [],
+        "numeric_equals": []
+    })
+    save_json(pack_dir / "expected.json", {
+        "note": "Optional gold output for regression tests. Populate when available."
+    })
+
+    # empty clarifications; you can fill interactively or with a UI later
+    save_json(pack_dir / "clarifications.json", {
+        "_note": "Populate with answers to Phase 1 clarification questions if you want batch runs."
+    })
+
+    return pack_dir