File size: 8,002 Bytes
caf8059
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Public PDF parsing contract and platform-owned fixtures, never supplied by the model."""
import base64
import json
import zlib
from pathlib import Path
from src.modules.execution.contracts import SandboxConfig, HttpCase
from src.modules.execution.pdf_dependency import FRAMEWORK, PROFILE, VERSION, COORDINATE, SHA512
from src.modules.skill.domain.models import Framework

SKILL_ID = 'pdf-java-development'
PDF_PROFILE = FRAMEWORK
PDF_VERSION = '0.1.0'
DEPENDENCIES = {COORDINATE: VERSION}
MAX_PDF_BYTES = 131072
MAX_BODY_BYTES = 180000
MAX_PAGES = 20
MAX_TEXT_BYTES = 24000

CONTRACT = '''Implement demo.App using Java17 HttpServer on port8080 and the provided Apache PDFBox3.0.8 jar (org.apache.pdfbox.Loader, PDDocument, PDFTextStripper). No other dependency/framework/download.
GET /health =>200 {"status":"ok"}. POST /api/pdf/analyze requires Authorization: Bearer demo-pdf, otherwise401 {"error":"UNAUTHORIZED"}.
Wrong method on either known endpoint =>405 {"error":"METHOD_NOT_ALLOWED"} before auth; unknown exact path=>404 {"error":"NOT_FOUND"}.
Analyze accepts UTF-8 JSON with exactly one field: {"base64":"<canonical padded standard Base64 PDF>"}, optional JSON whitespace. No URL/path/password fields or duplicate keys. Invalid/missing/extra fields, empty/noncanonical/invalid Base64 or query parameters=>400 {"error":"INVALID_INPUT"}.
Read at most180001 request bytes. Raw JSON>180000 or decoded PDF>131072 bytes=>413 {"error":"PDF_TOO_LARGE"}. Content-Type must start application/json (otherwise400 INVALID_INPUT).
PDF must begin %PDF- and parse with PDFBox; malformed=>400 {"error":"INVALID_PDF"}. Password-protected OR any encrypted PDF=>422 {"error":"ENCRYPTED_PDF"}, never decrypt/bypass. Zero pages=>400 INVALID_PDF; >20 pages=>422 {"error":"PAGE_LIMIT"}.
Use try-with-resources Loader.loadPDF(byte[]), reject document.isEncrypted(), PDFTextStripper with sortByPosition=true and start/end page, getText per page. Normalize CRLF/CR to LF and strip() surrounding whitespace; preserve internal line breaks. Count Unicode code points, not UTF16 units.
Success200 exactly {"pageCount":N,"pages":[{"page":1,"text":"..."},...],"totalCharacters":sumCodePoints,"status":"TEXT_EXTRACTED" or "NO_TEXT"}. NO_TEXT if every extracted page text is empty; do not pretend OCR. Keep blank pages. UTF8 text total>24000 or serialized response>60000 bytes=>413 {"error":"TEXT_LIMIT"}. Do not return partial success.
Use default single-thread executor or bounded executor, bounded memory input; never execute attachments, fetch URLs, use host filesystem, or log uploaded content. No external OCR. Close exchanges; JSON escaping must be valid. Files only src/main/java/demo/App.java and optional docs/README.md.'''


def framework_candidate():
    return Framework(profile_id=PDF_PROFILE, version=PDF_VERSION, dependencies=DEPENDENCIES, state='draft')


def fixture_pdf(texts, *, image=False, compressed=False):
    """Small deterministic authored PDFs with exact xref offsets; not harvested customer data."""
    objects = [b'<< /Type /Catalog /Pages 2 0 R >>', b'']
    objects.append(b'<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>')
    pages = []
    for text in texts:
        page_id = len(objects) + 1
        content_id = page_id + 1
        pages.append(page_id)
        content = b''
        if text:
            escaped = text.replace('\\', '\\\\').replace('(', '\\(').replace(')', '\\)')
            content = ('BT /F1 12 Tf 40 750 Td (' + escaped + ') Tj ET').encode('ascii')
        if image:
            # A genuine image-only page: raw one-pixel RGB image, with no text objects.
            content = b'q 100 0 0 100 40 600 cm /Im1 Do Q'
        resources = b'/Font << /F1 3 0 R >>'
        if image:
            resources += f' /XObject << /Im1 {content_id + 1} 0 R >>'.encode()
        objects.append(f'<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Resources << '.encode()+resources+f' >> /Contents {content_id} 0 R >>'.encode())
        payload = zlib.compress(content) if compressed else content
        objects.append(f'<< /Length {len(payload)}'.encode() + (b' /Filter /FlateDecode' if compressed else b'') + b' >>\nstream\n' + payload + b'\nendstream')
        if image:
            objects.append(b'<< /Type /XObject /Subtype /Image /Width 1 /Height 1 /ColorSpace /DeviceRGB /BitsPerComponent 8 /Length 3 >>\nstream\n\x00\x80\xff\nendstream')
    objects[1] = ('<< /Type /Pages /Count ' + str(len(pages)) + ' /Kids [' + ' '.join(str(p)+' 0 R' for p in pages) + '] >>').encode()
    data = bytearray(b'%PDF-1.4\n%\xe2\xe3\xcf\xd3\n')
    offsets = [0]
    for index, obj in enumerate(objects, 1):
        offsets.append(len(data)); data.extend(f'{index} 0 obj\n'.encode()+obj+b'\nendobj\n')
    xref = len(data)
    data.extend(f'xref\n0 {len(offsets)}\n0000000000 65535 f \n'.encode())
    for offset in offsets[1:]: data.extend(f'{offset:010} 00000 n \n'.encode())
    data.extend(f'trailer\n<< /Size {len(offsets)} /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n'.encode())
    return bytes(data)


def sample_pdf():
    return fixture_pdf(['Purchase order PO-2026-1003', 'Received 40 of 80 units.'])


def payload(pdf):
    return {'base64': base64.b64encode(pdf).decode()}


def scenario(image):
    health = HttpCase(path='/health', status=200, expected_json={'status':'ok'})
    cases = [health]
    def add(body, expected, status=200, method='POST', auth=True, path='/api/pdf/analyze'):
        cases.append(HttpCase(path=path, method=method, body=body, status=status, expected_json=expected,
                              headers={'Authorization':'Bearer demo-pdf'} if auth else {}))
    texts = ['Purchase order PO-2026-1003', 'Received 40 of 80 units.']
    add(payload(sample_pdf()), {'pageCount':2,'pages':[{'page':i+1,'text':text} for i,text in enumerate(texts)],
                                'totalCharacters':sum(map(len,texts)), 'status':'TEXT_EXTRACTED'})
    add(payload(fixture_pdf([''])), {'pageCount':1,'pages':[{'page':1,'text':''}],'totalCharacters':0,'status':'NO_TEXT'})
    add(payload(fixture_pdf([''],image=True)), {'pageCount':1,'pages':[{'page':1,'text':''}],'totalCharacters':0,'status':'NO_TEXT'})
    add(payload(fixture_pdf(['Compressed PDF text.'],compressed=True)), {'pageCount':1,'pages':[{'page':1,'text':'Compressed PDF text.'}],'totalCharacters':20,'status':'TEXT_EXTRACTED'})
    encrypted = json.loads(Path(__file__).with_name('pdf_fixtures.json').read_text())
    for key in ('password_required', 'empty_user_password'):
        add({'base64':encrypted[key]}, {'error':'ENCRYPTED_PDF'},422)
    for body in ({}, {'base64':''}, {'base64':'***'}, {'base64':42}, {'base64':'Zg'},
                 dict(payload(sample_pdf()), url='https://example.com/remote.pdf')):
        add(body, {'error':'INVALID_INPUT'},400)
    add(payload(b'not a pdf'), {'error':'INVALID_PDF'},400)
    add(payload(b'%PDF-1.4\nnot valid\n%%EOF'), {'error':'INVALID_PDF'},400)
    add(payload(fixture_pdf([''] * 21)), {'error':'PAGE_LIMIT'},422)
    add(payload(b'%PDF-' + b'x' * MAX_PDF_BYTES), {'error':'PDF_TOO_LARGE'},413)
    add(payload(fixture_pdf(['A'*24001],compressed=True)), {'error':'TEXT_LIMIT'},413)
    add(payload(sample_pdf()), {'error':'UNAUTHORIZED'},401,auth=False)
    add(None, {'error':'METHOD_NOT_ALLOWED'},405,method='GET')
    add({}, {'error':'NOT_FOUND'},404,path='/api/pdf/other')
    add(payload(sample_pdf()), {'error':'INVALID_INPUT'},400,path='/api/pdf/analyze?url=x')
    # Long upload proves HTTP-body handling independently of the preview UART transport.
    padded = sample_pdf() + b'\n%' + b' ' * 12000
    add(payload(padded), {'pageCount':2,'pages':[{'page':i+1,'text':text} for i,text in enumerate(texts)],
                         'totalCharacters':sum(map(len,texts)), 'status':'TEXT_EXTRACTED'})
    return SandboxConfig(image=image, skill_ids=[SKILL_ID], api_contract=CONTRACT, command_seconds=120,
                         dependency_profile=PROFILE, dependency_sha512=SHA512, tests=cases, health=health)