#!/usr/bin/env python3 """ cldr_dynamic_prompter.py Print a short, ticket-specific LLM prompt for a CLDR JIRA ticket. Now supports --template to load the prompt from a separate file. """ import re import sys import argparse from pathlib import Path from typing import Optional, Dict, List from jira import JIRA from cldr_ticket_reader import get_ticket_details_string # ========================= # CONFIG: paste credentials # ========================= JIRA_SERVER = "https://unicode-org.atlassian.net" JIRA_USER_EMAIL = "YOUR MAIL !!!" # <-- your email JIRA_API_TOKEN = "JIRA API KEY" # <-- your API token # optional; only needed if you use --auto-category OPENAI_API_KEY = "OPEN AI API KEY" # keep empty if not using auto-category OPENAI_MODEL = "gpt-4o-mini" CATEGORIES = ("Data Accuracy", "Documentation Issue", "Software Bug", "Feature Request") # Default template path (relative to this file) DEFAULT_TEMPLATE = Path(__file__).parent / "templates" / "phase1_prompt.md" # ----------------------------- # Jira + parsing helpers # ----------------------------- def get_jira_client_from_config() -> JIRA: if not JIRA_USER_EMAIL or not JIRA_API_TOKEN: raise RuntimeError("JIRA creds missing: set JIRA_USER_EMAIL and JIRA_API_TOKEN in cldr_dynamic_prompter.py") return JIRA(server=JIRA_SERVER, basic_auth=(JIRA_USER_EMAIL, JIRA_API_TOKEN)) def _get_match(pattern: str, text: str, flags=0, default: str = "") -> str: m = re.search(pattern, text, flags) return m.group(1).strip() if m else default def parse_report_text(report_text: str) -> Optional[Dict[str, str]]: """Extract fields from cldr_ticket_reader.py's report string.""" try: title = _get_match(r"^Title:\s*(.*)$", report_text, re.MULTILINE) reporter = _get_match(r"^Reporter:\s*(.*)$", report_text, re.MULTILINE, "N/A") priority = _get_match(r"^Priority:\s*(.*)$", report_text, re.MULTILINE, "N/A") components = _get_match(r"^Components:\s*(.*)$", report_text, re.MULTILINE, "None") labels = _get_match(r"^Labels:\s*(.*)$", report_text, re.MULTILINE, "None") desc = _get_match(r"Description:\s*\n---\n(.*?)\n---", report_text, re.DOTALL, "") code_blocks = re.findall(r"\{code.*?\}(.*?)\{code\}", desc, re.DOTALL) has_code = bool(code_blocks) links_block = _get_match(r"Connected Work Items:\n(.*?)\n\nDescription:", report_text, re.DOTALL, "") connected_items = [] if links_block: for ln in links_block.splitlines(): t = ln.strip().lstrip("-• ").strip() if t: connected_items.append(t) return { "title": title, "description": desc, "reporter": reporter, "priority": priority, "components": components, "labels": labels, "connected_items": connected_items, "has_code": has_code, } except Exception: return None def fetch_ticket_data(ticket_key: str, jira_client: JIRA) -> Dict[str, str]: """Use cldr_ticket_reader.get_ticket_details_string, then parse.""" report = get_ticket_details_string(ticket_key, jira_client) data = parse_report_text(report) if data: return data # fallback (rare) issue = jira_client.issue(ticket_key) title = getattr(issue.fields, "summary", "") or "" description = getattr(issue.fields, "description", "") or "" components = ", ".join([c.name for c in (issue.fields.components or [])]) or "None" labels = ", ".join(issue.fields.labels or []) or "None" reporter = getattr(getattr(issue.fields, "reporter", None), "displayName", "N/A") priority = getattr(getattr(issue.fields, "priority", None), "name", "N/A") has_code = bool(re.search(r"\{code.*?\}", description or "")) return { "title": title.strip(), "description": description.strip(), "reporter": reporter, "priority": priority, "components": components, "labels": labels, "connected_items": [], "has_code": has_code, } # ----------------------------- # Optional: auto category via LLM # ----------------------------- def auto_pick_category(title: str, description: str) -> str: if not OPENAI_API_KEY: return "Triage" try: from openai import OpenAI client = OpenAI(api_key=OPENAI_API_KEY) prompt = ( "Pick ONE category only from: " + ", ".join(f'"{c}"' for c in CATEGORIES) + ". Return just the category string.\n\n" f"Title: {title}\nDescription: {description}\n" ) resp = client.chat.completions.create( model=OPENAI_MODEL, messages=[{"role": "user", "content": prompt}], temperature=0, ) cat = (resp.choices[0].message.content or "").strip().strip('"') return cat if cat in CATEGORIES else "Triage" except Exception as e: print(f"[warn] Auto-category failed; using 'Triage': {e}", file=sys.stderr) return "Triage" # ----------------------------- # Topic detection (adds variety) # ----------------------------- def _norm_list_field(s: str) -> List[str]: if not s or s == "None": return [] return [x.strip().lower() for x in s.split(",") if x.strip()] def detect_topic(data: Dict[str, str]) -> str: title = (data.get("title") or "").lower() desc = (data.get("description") or "").lower() components = _norm_list_field(data.get("components")) labels = _norm_list_field(data.get("labels")) text = " ".join([title, desc] + components + labels) if any(k in text for k in ["intervalformatfallback", "datetime", "date time", "date-time", "skeleton", "pattern", "quotes", "apostrophe"]): return "Date/Time Patterns" if "icu" in text or any(k in labels for k in ["icu"]): return "ICU/Integration" if any(k in text for k in ["unit", "measurement", "unitpreference", "measurement system"]): return "Units" if any(k in text for k in ["plural", "pluralrules", "plural rules"]): return "Plural Rules" if any(k in text for k in ["collation", "sort order", "uca"]): return "Collation" if any(k in text for k in ["annotation", "emoji"]): return "Emoji/Annotations" if any(k in text for k in ["rbnf", "rule-based number", "spellout"]): return "RBNF" return "General/Locale Data" # ----------------------------- # Built-in prompt builders (fallback if no template) # ----------------------------- def build_triage_prompt(data: Dict[str, str]) -> str: return f"""You are assisting CLDR ticket triage. Ticket Title: {data['title']} Description: {data['description']} Context Reporter: {data['reporter']} | Priority: {data['priority']} Components: {data['components']} | Labels: {data['labels']} HasCodeBlock: {str(data['has_code']).lower()} Task Classify and return STRICT JSON with these keys: {{ "spam": true|false, "outOfScopeForCLDR": true|false, "needsEngineeringTC": true|false, "needsLanguageSpecialist": true|false, "stretchValidation": {{ "likelyTrue": true|false, "evidenceLinks": ["..."] }}, "changeTaskTypeToFixInSurveyTool": true|false, "notes": "1-3 sentences" }} Rules - spam: no meaningful info or advertisement. - outOfScopeForCLDR: e.g., emoji image design (not CLDR data). - Be concise; output ONLY valid JSON. """ def build_topic_prompt(category: str, topic: str, data: Dict[str, str]) -> str: base = f"""Analyze the ticket and answer succinctly. Ticket Title: {data['title']} Description: {data['description']} Context Reporter: {data['reporter']} | Priority: {data['priority']} Components: {data['components']} | Labels: {data['labels']} HasCodeBlock: {str(data['has_code']).lower()} Detected Category: {category} Topic: {topic} """ topic_asks = { "Date/Time Patterns": ( "Questions\n" "1) Which LDML/CLDR rule applies (cite section name)?\n" "2) Provide a corrected minimal example.\n" "3) Remove or escape curly quotes/apostrophes?\n" "4) Any ICU impact/doc reference?\n" "Return short bullets." ), "ICU/Integration": ( "Questions\n" "1) Which CLDR data groups ICU consumes here?\n" "2) What ICU does NOT focus on (if noted)?\n" "3) 2–4 bullets for doc outline + owners.\n" "Return short bullets." ), "Units": ( "Questions\n" "1) Which unit(s)/system?\n" "2) Preference, formatting, or conversion?\n" "3) One verification (locale + example).\n" "Return short bullets." ), "Plural Rules": ( "Questions\n" "1) Locales/categories?\n" "2) Logic vs examples?\n" "3) 1–2 minimal tests.\n" "Return short bullets." ), "Collation": ( "Questions\n" "1) Tailorings implicated?\n" "2) 3–5 sample order.\n" "3) Root vs locale tailoring?\n" "Return short bullets." ), "Emoji/Annotations": ( "Questions\n" "1) Annotation text vs image/design (image is out of scope)?\n" "2) Locales impacted?\n" "3) One-line next step.\n" "Return short bullets." ), "RBNF": ( "Questions\n" "1) Which ruleset(s)?\n" "2) Minimal before/after snippet.\n" "3) Example input + expected output.\n" "Return short bullets." ), "General/Locale Data": ( "Questions\n" "1) Which locales and data types?\n" "2) Is this accuracy, documentation, or feature scope?\n" "3) Concise next step (owner + artifact).\n" "Return short bullets." ), } return base + topic_asks.get(topic, topic_asks["General/Locale Data"]) # ----------------------------- # Template loading + rendering # ----------------------------- def load_template(path: Path) -> str: with open(path, "r", encoding="utf-8") as f: return f.read() def render_template(tmpl: str, mapping: Dict[str, str]) -> str: """Replace {{KEY}} tokens with mapping values (simple placeholder engine).""" def _sub(m): key = m.group(1).strip() return mapping.get(key, "") return re.sub(r"\{\{\s*([A-Z0-9_]+)\s*\}\}", _sub, tmpl) def build_from_template(ticket_key: str, template_path: Path, jira: Optional[JIRA]) -> str: if jira is not None: try: data = fetch_ticket_data(ticket_key, jira) except Exception as e: print(f"[warn] Jira fetch failed, rendering template with placeholders: {e}", file=sys.stderr) data = None else: data = None if data is None: # offline/placeholder mapping (still prints a prompt) mapping = { "TICKET_ID": ticket_key, "TITLE": "(unavailable: Jira error)", "DESCRIPTION": "(unavailable: Jira error)", "COMPONENTS": "None", "LABELS": "None", "REPORTER": "N/A", "PRIORITY": "N/A", "CONNECTED_ITEMS": "None", "HAS_CODE_BLOCK": "false", } else: connected = ", ".join(data["connected_items"]) if data.get("connected_items") else "None" mapping = { "TICKET_ID": ticket_key, "TITLE": data["title"], "DESCRIPTION": data["description"], "COMPONENTS": data["components"], "LABELS": data["labels"], "REPORTER": data["reporter"], "PRIORITY": data["priority"], "CONNECTED_ITEMS": connected, "HAS_CODE_BLOCK": str(data["has_code"]).lower(), } tmpl = load_template(template_path) return render_template(tmpl, mapping) # ----------------------------- # PUBLIC API for other tools # ----------------------------- def make_prompt(ticket_key: str, category: Optional[str] = None, auto_category: bool = False, template: Optional[str] = None) -> str: """ Return a ready-to-send prompt string for the given ticket. If a template file exists (explicit path or default), use it; otherwise fall back to built-ins. """ # Try to get Jira; do not crash if it fails. jira = None try: jira = get_jira_client_from_config() except Exception as e: print(f"[warn] Jira not available: {e}", file=sys.stderr) # Prefer template if provided or default exists template_path = Path(template) if template else DEFAULT_TEMPLATE if template_path.exists(): return build_from_template(ticket_key, template_path, jira) # Fallback to built-in prompts (requires Jira; will warn if missing) if jira is None: # Minimal fallback prompt if Jira is totally unavailable and no template is supplied return f"Ticket {ticket_key}: Jira unavailable and no template provided." data = fetch_ticket_data(ticket_key, jira) if category: chosen = category elif auto_category: chosen = auto_pick_category(data["title"], data["description"]) else: chosen = "Triage" if chosen == "Triage": return build_triage_prompt(data) topic = detect_topic(data) return build_topic_prompt(chosen, topic, data) # ----------------------------- # CLI (prints the prompt) # ----------------------------- def main(): parser = argparse.ArgumentParser( prog="cldr_dynamic_prompter", description="Print a ticket-specific LLM prompt for a CLDR JIRA ticket." ) parser.add_argument("ticket_key", help="e.g., CLDR-12345") parser.add_argument("--template", help="Path to a prompt template (default: templates/phase1_prompt.md)") parser.add_argument("--category", choices=list(CATEGORIES) + ["Triage"]) parser.add_argument("--auto-category", action="store_true") args = parser.parse_args() prompt = make_prompt(args.ticket_key, category=args.category, auto_category=args.auto_category, template=args.template) print(prompt) if __name__ == "__main__": main()