# retoor <retoor@molodetz.nl>
from __future__ import annotations
import re
from devplacepy . services . jobs . isslop . analysis . signals import (
AXIS_ORIGIN ,
AXIS_QUALITY ,
SEVERITY_MEDIUM ,
SEVERITY_STRONG ,
SEVERITY_WEAK ,
FileContext ,
Signal ,
)
PLACEHOLDER_PATTERNS : tuple [ tuple [ re . Pattern [ str ] , str ] , . . . ] = (
( re . compile ( r " your (code|logic|implementation) here " , re . IGNORECASE ) , " Placeholder body left in source " ) ,
( re . compile ( r " \ . \ . \ . ?rest of (the )?code " , re . IGNORECASE ) , " Truncated generation marker " ) ,
( re . compile ( r " \ . \ . \ . ?(existing|other|previous) code " , re . IGNORECASE ) , " Elided-code marker " ) ,
( re . compile ( r " / \ * \ s*implementation \ s* \ */ " , re . IGNORECASE ) , " Empty implementation stub " ) ,
( re . compile ( r " TODO:? implement( this)? \ b " , re . IGNORECASE ) , " Unimplemented TODO stub " ) ,
( re . compile ( r " TODO:? add (logic|implementation|code) " , re . IGNORECASE ) , " Unimplemented TODO stub " ) ,
( re . compile ( r " <!-- \ s*add content here \ s*--> " , re . IGNORECASE ) , " Placeholder HTML content " ) ,
( re . compile ( r " rest omitted " , re . IGNORECASE ) , " Truncated generation marker " ) ,
)
PROMPT_LEAK_PATTERNS : tuple [ re . Pattern [ str ] , . . . ] = (
re . compile ( r " as an ai (language )?model " , re . IGNORECASE ) ,
re . compile ( r " i (cannot|can ' t) (assist|help) with " , re . IGNORECASE ) ,
re . compile ( r " here( ' |’ " , re . IGNORECASE ) ,
re . compile ( r " \ bcertainly! \ s " , re . IGNORECASE ) ,
re . compile ( r " sure, here( ' s| is) " , re . IGNORECASE ) ,
re . compile ( r " i hope this helps " , re . IGNORECASE ) ,
re . compile ( r " let me know if (you|there) " , re . IGNORECASE ) ,
re . compile ( r " feel free to (adjust|modify|change|customize) " , re . IGNORECASE ) ,
re . compile ( r " note that this is a (simplified|basic) (example|implementation|version) " , re . IGNORECASE ) ,
re . compile ( r " in a (real|production)([- ]world)? (application|scenario|environment|setting) " , re . IGNORECASE ) ,
re . compile ( r " this is a placeholder " , re . IGNORECASE ) ,
re . compile ( r " replace . { 1,40} with your (actual|own) " , re . IGNORECASE ) ,
)
AI_SIGNATURE_PATTERNS : tuple [ re . Pattern [ str ] , . . . ] = (
re . compile ( r " generated (by|with) (chatgpt|copilot|claude|cursor|gemini|codeium|v0|bolt|ai) " , re . IGNORECASE ) ,
re . compile ( r " co-authored-by: \ s*(claude|copilot) " , re . IGNORECASE ) ,
re . compile ( r " \ U0001F916 generated with " ) ,
re . compile ( r " created with the help of ai " , re . IGNORECASE ) ,
re . compile ( r " \ bclaude code \ b " , re . IGNORECASE ) ,
re . compile ( r " powered by (chatgpt|gpt-4|claude|gemini) " , re . IGNORECASE ) ,
)
EMOJI_PATTERN = re . compile (
" [ \U0001F300 - \U0001FAFF ☀-➿⬀-⯿✅❌✨❗⭐] "
)
ENTHUSIASM_PATTERN = re . compile (
r " blazingly fast|seamlessly|robust and scalable|cutting[- ]edge|revolutioniz|effortlessly|supercharge|game[- ]chang " ,
re . IGNORECASE ,
)
LLM_LEXICON_PATTERN = re . compile (
r " \ b(delve|delving|delves|showcas(?:e|es|ing)|pivotal|intricate|meticulous(?:ly)?|realm "
r " |bolster(?:s|ing)?|garnered|underpins|seamless(?:ly)?|leverag(?:e|es|ing)|streamlin(?:e|es|ing) "
r " |elevat(?:e|es|ing)|holistic|robust and|comprehensive suite|game[- ]chang(?:er|ing) "
r " |it ' ?s (?:important|worth) (?:to note|noting)|keep in mind|as you can see|in this example) \ b " ,
re . IGNORECASE ,
)
LLM_LEXICON_THRESHOLD : int = 3
EM_DASH_THRESHOLD : int = 3
WORD_PATTERN = re . compile ( r " [a-z][a-z0-9]+ " )
NARRATION_MIN_OVERLAP : float = 0.55
NARRATION_MIN_TOKENS : int = 2
def _narration_signals ( context : FileContext ) - > list [ Signal ] :
findings : list [ Signal ] = [ ]
line_lookup = { number : text for number , text in context . comments }
for number , comment in context . comments :
if number + 1 in line_lookup or number > = len ( context . lines ) :
continue
comment_tokens = set ( WORD_PATTERN . findall ( comment . lower ( ) ) )
if len ( comment_tokens ) < NARRATION_MIN_TOKENS :
continue
code_line = context . lines [ number ] . lower ( )
code_tokens = set ( WORD_PATTERN . findall ( code_line ) )
if not code_tokens :
continue
overlap = len ( comment_tokens & code_tokens ) / len ( comment_tokens )
if overlap > = NARRATION_MIN_OVERLAP :
findings . append (
Signal (
code = " COMMENT_NARRATION " ,
title = " Comment narrates the next line of code " ,
severity = SEVERITY_MEDIUM ,
axis = AXIS_ORIGIN ,
weight = 2.0 ,
line = number ,
evidence = comment ,
)
)
return findings [ : 8 ]
def detect_textual ( context : FileContext ) - > list [ Signal ] :
findings : list [ Signal ] = [ ]
for number , raw in enumerate ( context . lines , start = 1 ) :
for pattern , title in PLACEHOLDER_PATTERNS :
if pattern . search ( raw ) :
findings . append (
Signal ( " PLACEHOLDER_COMMENT " , title , SEVERITY_STRONG , AXIS_QUALITY , 4.0 , number , raw . strip ( ) )
)
break
for pattern in AI_SIGNATURE_PATTERNS :
if pattern . search ( raw ) :
findings . append (
Signal (
" AI_SIGNATURE " ,
" AI tool self-attribution signature " ,
SEVERITY_STRONG ,
AXIS_ORIGIN ,
4.0 ,
number ,
raw . strip ( ) ,
)
)
break
for number , comment in context . comments :
for pattern in PROMPT_LEAK_PATTERNS :
if pattern . search ( comment ) :
findings . append (
Signal (
" PROMPT_LEAK " ,
" Assistant conversation leakage in committed source " ,
SEVERITY_STRONG ,
AXIS_QUALITY ,
4.0 ,
number ,
comment ,
)
)
break
emoji_hits = [ ( number , comment ) for number , comment in context . comments if EMOJI_PATTERN . search ( comment ) ]
if len ( emoji_hits ) > = 2 :
number , comment = emoji_hits [ 0 ]
findings . append (
Signal (
" EMOJI_COMMENTS " ,
f " Emoji used in { len ( emoji_hits ) } comments " ,
SEVERITY_MEDIUM ,
AXIS_ORIGIN ,
1.0 ,
number ,
comment ,
)
)
enthusiasm_hits = [ ( number , comment ) for number , comment in context . comments if ENTHUSIASM_PATTERN . search ( comment ) ]
if enthusiasm_hits :
number , comment = enthusiasm_hits [ 0 ]
findings . append (
Signal (
" MARKETING_TONE " ,
" Marketing-style enthusiasm in comments " ,
SEVERITY_WEAK ,
AXIS_QUALITY ,
1.0 ,
number ,
comment ,
)
)
comment_text = " " . join ( comment for _ , comment in context . comments )
lexicon_hits = { hit . lower ( ) if isinstance ( hit , str ) else hit for hit in LLM_LEXICON_PATTERN . findall ( comment_text ) }
if len ( lexicon_hits ) > = LLM_LEXICON_THRESHOLD :
findings . append (
Signal (
" LLM_LEXICON " ,
f " { len ( lexicon_hits ) } distinct LLM signature words in comments " ,
SEVERITY_MEDIUM ,
AXIS_ORIGIN ,
2.0 ,
context . comments [ 0 ] [ 0 ] if context . comments else 1 ,
" , " . join ( sorted ( str ( hit ) for hit in lexicon_hits ) [ : 6 ] ) ,
)
)
em_dashes = comment_text . count ( " \u2014 " )
if em_dashes > = EM_DASH_THRESHOLD :
findings . append (
Signal (
" EM_DASH_OVERUSE " ,
f " { em_dashes } em dashes in comments, a strong LLM prose habit " ,
SEVERITY_MEDIUM ,
AXIS_ORIGIN ,
1.0 ,
context . comments [ 0 ] [ 0 ] if context . comments else 1 ,
f " { em_dashes } em dashes " ,
)
)
findings . extend ( _narration_signals ( context ) )
return findings