Dynavera/apps/onboarding/consumers/prompts.py
2026-03-22 17:26:02 +00:00

197 lines
No EOL
11 KiB
Python

import json
__all__ = ['OnboardingPrompts']
class OnboardingPrompts:
@staticmethod
def default_system_prompt():
return (
"You are a helpful onboarding assistant that helps new employees get onboarded to their new company."
"You may use relevant tools to assist you to provide the best support."
)
@staticmethod
def force_reasoning_prompt():
return "Double check your reasoning and provide the final improved answer."
@staticmethod
def curriculum_generation_prompt(role_uuid: str, role_name: str, initial_context: str = '') -> str:
context_section = (
f"\nPrimary source — retrieved training documents for this role:\n{initial_context}\n"
if initial_context else
"\nNo training documents were retrieved. Module titles must be based solely on what you find via tools.\n"
)
return (
f"Create an onboarding curriculum for the '{role_name}' role (role_uuid: {role_uuid}).\n"
f"{context_section}\n"
"The retrieved documents above are your primary source. "
"Module titles MUST reflect the specific topics, responsibilities, and competencies described in those documents — "
"do NOT default to generic onboarding titles (e.g. 'Orientation', 'IT Setup') unless the documents support them.\n"
"You may call tools to supplement your understanding:\n"
"- Call get_role_context if you need the role description\n"
"- Call search_knowledge with a more specific query if the primary source is insufficient for a particular area\n"
"Decide how many modules are appropriate for this role's complexity — up to 15. "
"Output ONLY a valid JSON array of strings representing module titles. "
"Example: [\"Introduction\", \"Safety\", \"Operations\"]"
)
@staticmethod
def knowledge_generation_prompt(topic, context_markdown):
return (
f"Write a practical onboarding training guide for the topic '{topic}'. "
"Think step-by-step internally before writing the final answer. "
"Use the MCP search context below as your primary source, and call additional tools if needed. "
"If no indexed documents are available, provide a concise best-practice overview and clearly say no indexed documents were found. "
"Use Markdown formatting and do NOT include a table of contents in this section. "
"Generate substantial depth: target 900-1400 words. "
"Include these sections in order: Overview, Core Concepts, Role-Specific Workflow, Practical Examples, Common Pitfalls, and Action Checklist. "
"In Practical Examples, provide at least 2 concrete examples relevant to this role/topic. "
"In Action Checklist, provide at least 8 actionable checklist items.\n\n"
f"Topic: {topic}\n"
f"MCP search context:\n{context_markdown}"
)
@staticmethod
def quiz_generation_prompt(question_count, module_briefs):
return (
"Create a final onboarding quiz that assesses all generated modules. "
f"Output ONLY a valid JSON array of exactly {question_count} question objects. "
"Use a mix of question types: at least 2 short-answer questions and at least 2 multiple-choice questions. "
"For multiple-choice objects: field_type='select', options (4 unique strings), and validation.correct_option. "
"For short-answer objects: field_type='textarea' (or 'text') and validation.accepted_answers (array of valid answers/keywords). "
"Each object MUST include key, label, field_type, required=true, and validation.explanation. "
"Cover all topics with balanced difficulty and avoid ambiguous wording.\n\n"
f"Modules JSON:\n{json.dumps(module_briefs, ensure_ascii=False)}"
)
@staticmethod
def quiz_generation_retry_prompt(question_count, module_briefs):
return OnboardingPrompts.quiz_generation_prompt(question_count, module_briefs) + (
"Return ONLY raw JSON. Do not use markdown fences. Do not include explanations outside JSON."
)
@staticmethod
def progress_monitoring_prompt(progress_context):
return (
"You are a progress monitoring agent for onboarding. "
"Analyze the role onboarding data below and provide concise feedback with:\n"
"1) current status\n2) strengths\n3) gaps\n4) next actions\n"
"Use prior learner question/answer evidence and any saved marking details when available. "
"If evidence is insufficient, explicitly state what is missing.\n"
"Keep it short and practical.\n\n"
f"Progress context JSON:\n{json.dumps(progress_context)}"
)
### Default agent system prompts (canonical source of truth) ###
@staticmethod
def default_curriculum_prompt(role_name: str) -> str:
return (
f"You are an instructional design assistant for onboarding the role '{role_name}'. "
"Your job is to teach the learner what the role does and how responsibilities are performed in practice. "
"Create a structured curriculum with clear objectives, prerequisite knowledge, core competencies, "
"hands-on tasks, and measurable outcomes. Avoid role-play and avoid claiming to be in the role; "
"focus on teaching the role responsibilities, expected decisions, and quality standards."
)
@staticmethod
def default_knowledge_prompt(role_name: str) -> str:
return (
f"You are a domain knowledge tutor for the role '{role_name}'. "
"Answer questions with concise explanations, practical examples, and references to expected workflows. "
"When possible, explain why a step matters, common mistakes, and how to verify correctness. "
"Do not act as the role holder; teach the learner how to perform the role responsibly and accurately."
)
@staticmethod
def default_assessment_prompt(role_name: str) -> str:
return (
f"You are an assessment designer for onboarding the role '{role_name}'. "
"Generate scenario-based checks that evaluate conceptual understanding, decision-making, and execution quality. "
"Include rubrics, expected evidence, and feedback that explains gaps and remediation steps. "
"Assess against role responsibilities and standards, not generic trivia."
)
@staticmethod
def default_monitor_prompt(role_name: str) -> str:
return (
f"You are a progress coaching assistant for learners training for the role '{role_name}'. "
"Track competency milestones, summarize strengths and weaknesses, and recommend next actions. "
"Flag unresolved risks, missing evidence, and topics requiring revision. "
"Keep feedback specific, actionable, and tied to role responsibilities and expected outcomes."
)
@staticmethod
def refine_curriculum_prompt(role_name: str, base_prompt: str, document_text: str) -> str:
return (
f"You are refining a curriculum agent's system prompt for the '{role_name}' role. "
"Training documents have been uploaded. Rewrite the system prompt below so it incorporates "
"the specific topics and subject matter from those documents. "
"Preserve all original instructions and add concrete topic guidance where relevant. "
"Return ONLY the refined system prompt text — no commentary, no labels.\n\n"
f"Original system prompt:\n{base_prompt}\n\n"
f"Training document content:\n{document_text}"
)
@staticmethod
def refine_assessment_prompt(role_name: str, base_prompt: str, document_text: str) -> str:
return (
f"You are refining an assessment agent's system prompt for the '{role_name}' role. "
"Training documents have been uploaded. Rewrite the system prompt below so it targets "
"the core competency areas and standards described in those documents. "
"Focus on what should be assessed — key responsibilities, decision points, and quality criteria — "
"not on topic lists. Preserve all original instructions. "
"Return ONLY the refined system prompt text — no commentary, no labels.\n\n"
f"Original system prompt:\n{base_prompt}\n\n"
f"Training document content:\n{document_text}"
)
FALLBACK_SYSTEM_PROMPT = 'You are a helpful onboarding assistant.'
KA_HELP_FALLBACK = (
"I couldn't reach the knowledge model right now. "
"Please try again, or clarify which part of this module is confusing and I can provide a shorter explanation."
)
@staticmethod
def grading_prompt(ai_fields, page_responses):
return (
'You are grading a completed onboarding final quiz. '
'Evaluate each learner answer for correctness using the question prompt and validation hints. '
'Do NOT grade multiple-choice select questions here; they are graded separately. '
'Grade only the provided non-select questions (for example short-answer/textarea). '
'For short-answer questions, use validation.accepted_answers semantically and allow equivalent phrasing. '
'For incorrect answers, provide a brief coaching reason that explains what is missing or incorrect, '
'but DO NOT reveal the correct answer, exact option text, or accepted-answer phrases. '
'Keep each reason to one short sentence. '
'Return ONLY JSON object with keys: correct_count (int), gradable_count (int), per_question (array of '
'{key, correct, reason}). Do not include markdown.'
f"\n\nQuiz fields JSON:\n{json.dumps(ai_fields, ensure_ascii=False)}"
f"\n\nLearner answers JSON:\n{json.dumps(page_responses, ensure_ascii=False)}"
)
@staticmethod
def ka_help_prompt(role_name, page_title, page_body, user_message):
return (
"Help the learner understand this onboarding page. Keep the explanation concise and practical. "
"Use markdown with bullets when useful.\n\n"
f"Role: {role_name}\n"
f"Page Title: {page_title}\n"
f"Page Body (excerpt): {str(page_body)[:2000]}\n"
f"Learner question: {user_message}"
)
@staticmethod
def ka_page_revision_prompt(role_name, page_title, page_body, user_message):
return (
"Revise the onboarding page content by integrating the learner's clarification request directly into the main page text. "
"Use the current page as the source of truth, preserve useful structure, and improve clarity and examples where needed. "
"Do not append a separate 'Clarification' section. "
"Return ONLY the fully revised markdown page body. "
"When you have finished the revision, write [END] on its own line and stop.\n\n"
f"Role: {role_name}\n"
f"Page Title: {page_title}\n"
f"Learner clarification request: {user_message}\n\n"
f"Current page markdown:\n{str(page_body)[:12000]}"
)