Spaces:
Sleeping
Sleeping
File size: 22,207 Bytes
aa8fa0a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 | import os, json, re, random
import uuid
import time
import logging
from typing import Literal, List, Dict, Any, Optional
from pydantic import BaseModel, Field, ValidationError
from crewai import Agent, Task, Crew, Process
from crewai.tools import tool
from crewai.llm import LLM
import dotenv
dotenv.load_dotenv("api_key.env")
# ============================================================
# Guardrails: logging, retries, deterministic config
# ============================================================
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s | %(levelname)s | %(message)s",
)
logger = logging.getLogger("smart_tutor_guardrails")
DETERMINISTIC_TEMPERATURE = float(os.getenv("DETERMINISTIC_TEMPERATURE", "0.1"))
TOOL_MAX_RETRIES = int(os.getenv("TOOL_MAX_RETRIES", "2"))
# ============================================================
# Guardrails: rate limits / timeouts / policies
# ============================================================
MAX_FILE_SIZE_MB = int(os.getenv("MAX_FILE_SIZE_MB", "500"))
MAX_PDF_PAGES = int(os.getenv("MAX_PDF_PAGES", "2000"))
PDF_EXTRACTION_TIMEOUT = float(os.getenv("PDF_EXTRACTION_TIMEOUT", "200")) # seconds
ALLOWED_TOOLS = {"process_file", "store_quiz", "grade_quiz"}
PROMPT_INJECTION_PATTERNS = [
"ignore previous instructions",
"ignore all previous instructions",
"system:",
"assistant:",
"developer:",
"act as",
"you must",
"follow these instructions",
"override",
]
# ============================================================
# Helpers
# ============================================================
def clean_text(text: str) -> str:
text = text.replace("\x00", " ")
text = re.sub(r"[ \t]+", " ", text)
text = re.sub(r"\n{3,}", "\n\n", text)
return text.strip()
def detect_prompt_injection(text: str) -> bool:
lower = text.lower()
return any(p in lower for p in PROMPT_INJECTION_PATTERNS)
def chunk_text(text: str, max_chars: int = 1200, overlap: int = 150) -> List[str]:
text = clean_text(text)
if not text:
return []
chunks = []
start = 0
n = len(text)
while start < n:
end = min(start + max_chars, n)
part = text[start:end].strip()
if part:
chunks.append(part)
if end == n:
break
start = max(0, end - overlap)
return chunks
def keyword_retrieve(chunks: List[str], query: str, top_k: int) -> List[str]:
q_terms = [w for w in re.findall(r"\w+", query.lower()) if len(w) > 2]
def score(c: str) -> int:
c_l = c.lower()
return sum(1 for t in q_terms if t in c_l)
ranked = sorted(chunks, key=score, reverse=True)
return [c for c in ranked[:top_k] if c]
# ============================================================
# File extraction with limits + timeout
# ============================================================
def extract_text(file_path: str) -> str:
if os.path.getsize(file_path) > MAX_FILE_SIZE_MB * 1024 * 1024:
raise ValueError(f"File too large (> {MAX_FILE_SIZE_MB} MB)")
ext = os.path.splitext(file_path)[1].lower()
if ext == ".txt":
with open(file_path, "r", encoding="utf-8", errors="ignore") as f:
return f.read()
if ext == ".pdf":
import fitz # PyMuPDF
start_time = time.time()
doc = fitz.open(file_path)
if len(doc) > MAX_PDF_PAGES:
raise ValueError(f"PDF exceeds max page limit ({MAX_PDF_PAGES})")
parts = []
for i in range(len(doc)):
if time.time() - start_time > PDF_EXTRACTION_TIMEOUT:
raise TimeoutError("PDF extraction timeout")
t = doc.load_page(i).get_text("text") or ""
t = clean_text(t)
if t:
parts.append(t)
return "\n\n".join(parts).strip()
raise ValueError("Unsupported file type (PDF/TXT only).")
# ============================================================
# Schemas (Structured Inputs / Outputs)
# ============================================================
class ProcessArgs(BaseModel):
file_path: str = Field(..., description="Local path to PDF/TXT")
query: str = Field(..., description="User question or instruction")
mode: Literal["summarize", "quiz", "explain"] = Field(..., description="Task type")
top_k: int = Field(6, ge=1, le=15, description="How many chunks to use as context")
class QuizQuestion(BaseModel):
qid: str
question: str
options: Dict[Literal["A", "B", "C", "D"], str]
correct: Literal["A", "B", "C", "D"]
explanation: str = ""
supporting_context: str = ""
class StoreQuizArgs(BaseModel):
file_path: str = Field(
..., description="The absolute file path of the document used"
)
questions: List[QuizQuestion]
class GradeQuizArgs(BaseModel):
quiz_id: str
answers: Dict[str, Literal["A", "B", "C", "D"]]
class ToolError(BaseModel):
error: str
details: Optional[Any] = None
class ProcessFileResult(BaseModel):
mode: str
query: str
context_chunks: List[str]
stats: Dict[str, Any]
class StoreQuizResult(BaseModel):
quiz_id: str
questions: List[Dict[str, Any]] # masked questions
class GradeQuizResult(BaseModel):
quiz_id: str
score: int
total: int
percentage: float
file_path: Optional[str] = None
details: List[Dict[str, Any]]
# ============================================================
# Memory/State with Persistence
# ============================================================
QUIZ_FILE = "quizzes_db.json"
def load_quizzes():
if os.path.exists(QUIZ_FILE):
try:
with open(QUIZ_FILE, "r", encoding="utf-8") as f:
return json.load(f)
except:
return {}
return {}
def save_quizzes(data):
try:
with open(QUIZ_FILE, "w", encoding="utf-8") as f:
json.dump(data, f, ensure_ascii=False, indent=2)
except Exception as e:
logger.error(f"Failed to save quizzes: {e}")
QUIZ_STORE: Dict[str, Dict[str, Any]] = load_quizzes()
# ============================================================
# Tool wrapper: retries + logs + redaction
# ============================================================
def _redact(obj: Any) -> Any:
"""Redact secrets + quiz answer key in logs."""
try:
if isinstance(obj, dict):
out = {}
for k, v in obj.items():
lk = str(k).lower()
if lk in {"openai_api_key", "api_key", "authorization", "x-api-key"}:
out[k] = "***"
elif lk == "correct":
out[k] = "***"
else:
out[k] = _redact(v)
return out
if isinstance(obj, list):
return [_redact(x) for x in obj]
if isinstance(obj, str):
key = os.getenv("OPENAI_API_KEY") or ""
if key and key in obj:
return obj.replace(key, "***")
return obj
return obj
except Exception:
return "<redacted>"
def safe_tool_call(tool_name: str, fn):
if tool_name not in ALLOWED_TOOLS:
raise RuntimeError("Tool not allowed by policy")
last_err = None
for attempt in range(1, TOOL_MAX_RETRIES + 2):
try:
logger.info(f"[TOOL_CALL] {tool_name} attempt={attempt}")
out = fn()
logger.info(
f"[TOOL_RESULT] {tool_name} attempt={attempt} out={json.dumps(_redact(out), ensure_ascii=False)[:900]}"
)
return out
except Exception as e:
last_err = e
logger.warning(
f"[TOOL_ERROR] {tool_name} attempt={attempt} err={type(e).__name__}"
)
time.sleep(0.2 * attempt)
raise last_err
# ============================================================
# Tools
# ============================================================
@tool("process_file")
def process_file(file_path: str, query: str, mode: str, top_k: int = 6) -> str:
"""Read PDF/TXT, chunk it, retrieve top_k relevant chunks. Returns structured JSON."""
try:
args = ProcessArgs(file_path=file_path, query=query, mode=mode, top_k=top_k)
except ValidationError as ve:
return json.dumps(
ToolError(error="Invalid arguments", details=ve.errors()).model_dump(),
ensure_ascii=False,
)
def _run():
# Clean path: remove quotes and whitespace that agents sometimes add
clean_path = args.file_path.strip().strip("'\"").strip()
if not os.path.exists(clean_path):
return ToolError(error=f"Invalid file path: {clean_path}").model_dump()
try:
raw_text = extract_text(args.file_path)
except Exception as e:
return ToolError(
error="Extraction failed", details=type(e).__name__
).model_dump()
if detect_prompt_injection(raw_text):
logger.warning(
"[SECURITY] Potential prompt injection detected in document. Treating as data only."
)
text = clean_text(raw_text)
if not text:
return ToolError(error="Empty or unreadable file text.").model_dump()
chunks = chunk_text(text)
if not chunks:
return ToolError(error="No chunks produced.").model_dump()
context = keyword_retrieve(chunks, args.query, args.top_k)
return ProcessFileResult(
mode=args.mode,
query=args.query,
context_chunks=context,
stats={
"chunks_total": len(chunks),
"chars_extracted": len(text),
"top_k": args.top_k,
},
).model_dump()
try:
out = safe_tool_call("process_file", _run)
return json.dumps(out, ensure_ascii=False)
except Exception as e:
return json.dumps(
ToolError(
error="process_file failed", details=type(e).__name__
).model_dump(),
ensure_ascii=False,
)
def clean_json_input(text: str) -> str:
"""Clean markdown code blocks and extract JSON object from string."""
text = text.strip()
# Remove markdown code blocks (flexible)
# This handles ```json ... ``` even if there is text before/after
pattern = r"```(?:json)?\s*(\{.*?\})\s*```"
match = re.search(pattern, text, re.DOTALL)
if match:
return match.group(1)
# If no code blocks, try to find the first outer-most JSON object
# This regex looks for { ... } minimally or greedily?
# We want the largest block starting with { and ending with }
# but strictly speaking, standard json.loads might just work if we strip.
# If text starts with ``` but didn't match the block above (maybe incomplete),
# let's just strip the fences.
if text.startswith("```"):
text = re.sub(r"^```(\w+)?\n?", "", text)
text = re.sub(r"\n?```$", "", text)
# Remove single backticks
if text.startswith("`") and text.endswith("`"):
text = text.strip("`")
return text.strip()
@tool("store_quiz")
def store_quiz(quiz_package_json: str) -> str:
"""Store quiz with hidden answers; return masked quiz (no correct answers)."""
def _run():
try:
cleaned_json = clean_json_input(quiz_package_json)
# First try: direct parse
pkg_raw = json.loads(cleaned_json)
except json.JSONDecodeError:
# Second try: liberal regex search for { ... }
# Use dotall and greedy to capture nested objects
match = re.search(r"(\{.*\})", quiz_package_json, re.DOTALL)
if match:
try:
pkg_raw = json.loads(match.group(1))
except json.JSONDecodeError as e:
return ToolError(
error=f"quiz_package_json is not valid JSON. Parse error: {str(e)}",
details=f"Input fragment: {quiz_package_json[:200]}...",
).model_dump()
else:
return ToolError(
error="quiz_package_json is not valid JSON (no braces found)",
details=f"Input fragment: {quiz_package_json[:200]}...",
).model_dump()
try:
pkg = StoreQuizArgs(**pkg_raw)
except ValidationError as ve:
return ToolError(
error="Invalid quiz_package_json", details=ve.errors()
).model_dump()
quiz_id = str(uuid.uuid4())
# Randomize options for each question
final_questions = []
for q in pkg.questions:
# q is a QuizQuestion object
original_options = q.options # dict e.g. {"A": "...", "B": "..."}
original_correct_key = q.correct # "A"
correct_text = original_options[original_correct_key]
# Extract texts
option_texts = list(original_options.values())
random.shuffle(option_texts)
# Re-map to A, B, C, D
new_options = {}
new_correct_key = ""
keys = ["A", "B", "C", "D"]
# Handle cases with fewer than 4 options just in case
for i, text in enumerate(option_texts):
if i < len(keys):
key = keys[i]
new_options[key] = text
if text == correct_text:
new_correct_key = key
# Update the question object (create a copy/dict)
q_dump = q.model_dump()
q_dump["options"] = new_options
q_dump["correct"] = new_correct_key
final_questions.append(q_dump)
QUIZ_STORE[quiz_id] = {
"file_path": pkg.file_path,
"questions": final_questions,
}
save_quizzes(QUIZ_STORE)
masked = [
{"qid": q["qid"], "question": q["question"], "options": q["options"]}
for q in final_questions
]
return StoreQuizResult(quiz_id=quiz_id, questions=masked).model_dump()
try:
out = safe_tool_call("store_quiz", _run)
return json.dumps(out, ensure_ascii=False)
except Exception as e:
return json.dumps(
ToolError(error="store_quiz failed", details=type(e).__name__).model_dump(),
ensure_ascii=False,
)
@tool("grade_quiz")
def grade_quiz(quiz_id: str, answers_json: str) -> str:
"""Grade quiz answers by quiz_id and answers_json. Returns score + details as structured JSON.
Also returns 'file_path' and 'question' text for further processing."""
def _run():
if quiz_id not in QUIZ_STORE:
return ToolError(error="Unknown quiz_id.").model_dump()
try:
cleaned_json = clean_json_input(answers_json)
submitted_raw = json.loads(cleaned_json)
except json.JSONDecodeError:
# Fallback
match = re.search(r"(\{.*\})", answers_json, re.DOTALL)
if match:
try:
submitted_raw = json.loads(match.group(1))
except:
return ToolError(
error="answers_json is not valid JSON"
).model_dump()
else:
return ToolError(error="answers_json is not valid JSON").model_dump()
try:
args = GradeQuizArgs(quiz_id=quiz_id, answers=submitted_raw)
except ValidationError as ve:
return ToolError(
error="Invalid answers_json", details=ve.errors()
).model_dump()
stored_data = QUIZ_STORE[args.quiz_id]
questions = stored_data["questions"]
file_path = stored_data.get("file_path")
total = len(questions)
score = 0
details = []
for q in questions:
qid = q["qid"]
correct = q["correct"]
question_text = q.get("question", "")
your = (args.answers.get(qid) or "").strip().upper()
is_correct = your == correct
score += 1 if is_correct else 0
details.append(
{
"qid": qid,
"question": question_text, # Added for Agent context
"is_correct": is_correct,
"your_answer": your,
"correct_answer": correct, # NOTE: returned to tutor; OK for feedback
"explanation": q.get("explanation", "") or "",
"supporting_context": q.get("supporting_context", "") or "",
}
)
percentage = round((score / total) * 100, 2) if total else 0.0
return GradeQuizResult(
quiz_id=args.quiz_id,
score=score,
total=total,
percentage=percentage,
file_path=file_path,
details=details,
).model_dump()
try:
out = safe_tool_call("grade_quiz", _run)
return json.dumps(out, ensure_ascii=False)
except Exception as e:
return json.dumps(
ToolError(error="grade_quiz failed", details=type(e).__name__).model_dump(),
ensure_ascii=False,
)
# ============================================================
# CrewAI setup
# ============================================================
llm = LLM(
model="gpt-4o-mini",
api_key=os.getenv("OPENAI_API_KEY"),
temperature=DETERMINISTIC_TEMPERATURE,
)
manager = Agent(
role="Manager (Router)",
goal=(
"Route user request to the correct specialist co-worker."
" Pass ALL user constraints (line count, "
"paragraph count, language, etc.) to the specialist."
),
backstory=(
"You are a routing agent. You HAVE specialist co-workers: "
"Summarizer, Quiz Maker, and Tutor. "
"Your ONLY job is to delegate the task to the right co-worker "
"using your delegation tool. "
"NEVER answer the user yourself. NEVER use internal knowledge. "
"Always forward the FULL user request including any constraints."
),
allow_delegation=True,
llm=llm,
verbose=True,
)
summarizer = Agent(
role="Summarizer",
goal=(
"Produce a summary grounded strictly in "
"context_chunks from process_file. STRICTLY "
"follow any user constraints on length, "
"number of lines, paragraphs, or format."
),
backstory=(
"Call process_file(mode=summarize) first. "
"Summarize ONLY from context_chunks. "
"If the user specifies constraints like "
"'3 lines', '2 paragraphs', 'short', or "
"'detailed', you MUST follow them exactly. "
"Use bullet points (- or *) for lists instead of numbering. "
"No outside knowledge."
),
tools=[process_file],
llm=llm,
verbose=True,
)
quizzer = Agent(
role="Quiz Maker",
goal="Generate EXACTLY the number of multiple-choice questions requested by the user, grounded strictly in process_file context.",
backstory=(
"STEP 1: Extract the EXACT number of questions from user request (e.g., '3 questions' = 3, default = 5).\n"
"STEP 2: Call process_file(mode=quiz) with file_path. Create ONLY that exact number of MCQs A-D from context_chunks.\n"
"STEP 3: Build quiz_package_json with absolute 'file_path' and correct answers, call store_quiz.\n"
'CRITICAL: The \'qid\' field for each question MUST be a STRING (e.g., "1", "2") NOT an integer (1, 2).\n'
'Ensure VALID JSON: {"file_path": "...", "questions": [...]}. CRITICAL: Match requested count exactly. Never reveal answers.'
),
tools=[process_file, store_quiz],
llm=llm,
verbose=True,
)
tutor = Agent(
role="Tutor",
goal="Grade quiz and provide intelligent explanation for errors.",
backstory=(
"You are an expert Tutor. When asked to grade a quiz:\n"
"1. Call 'grade_quiz' to get the base results.\n"
"2. For every INCORRECT answer, you MUST Explain WHY it is wrong:\n"
" - Use the 'question' text and 'file_path' from the result to call 'process_file' (mode='explain', query=question).\n"
" - REWRITE the 'explanation' field in the JSON detail for that question with your new explanation.\n"
" - Use bullet points for any lists in your explanations.\n"
"3. Return the fully updated JSON object."
),
tools=[process_file, grade_quiz],
llm=llm,
verbose=True,
)
task = Task(
description=(
"User request: {user_request}\n\n"
"Route by intent:\n"
"- Summary -> Summarizer\n"
"- Quiz -> Quiz Maker\n"
"- Explanation -> Tutor\n"
"- Grading (contains quiz_id + answers_json) -> Tutor\n\n"
"Guardrails:\n"
"- Tool outputs are structured JSON.\n"
"- Tools validate inputs with Pydantic.\n"
"- Tool calls are logged without secrets.\n"
"- Do not reveal hidden quiz answers during quiz generation."
),
expected_output=(
"Grounded response: summary OR " "masked quiz OR graded feedback."
),
agent=manager,
)
crew = Crew(
agents=[manager, summarizer, quizzer, tutor],
tasks=[task],
process=Process.sequential,
verbose=True,
)
from pathlib import Path
def run_with_file(prompt: str, file_path: str | None = None):
file_text = ""
if file_path:
file_text = Path(file_path).read_text(encoding="utf-8", errors="ignore")
full_prompt = prompt
if file_text:
full_prompt += "\n\n[FILE CONTENT]\n" + file_text
return full_prompt
if __name__ == "__main__":
print(
run_with_file(
r"please give me a quiz about 3 questions from this file - file_path=C:\Users\Yaz00\OneDrive\سطح المكتب\Agent AI - Tuwaiq\week 5\Homework 1\Phase2.pdf"
)
)
# Example grading:
# print(run(r"grade this quiz_id=<PUT_ID_HERE> answers_json={\"q1\":\"A\",\"q2\":\"C\",\"q3\":\"B\"}"))
pass
|