Skip to content

armgit / kong

Public Imported from GitHub
0

The world's first agentic reverse engineer.

fix some circular import bugs

Amruth9724f874b1f6
diff --git a/kong/agent/deobfuscator.py b/kong/agent/deobfuscator.py
index d123dfe..faadf01 100644
--- a/kong/agent/deobfuscator.py
+++ b/kong/agent/deobfuscator.py
@@ -10,13 +10,13 @@ from enum import Enum
from pathlib import Path
from typing import TYPE_CHECKING
-from kong.agent.analyzer import AnalysisContext, LLMClient, LLMResponse
from kong.agent.models import FunctionResult
from kong.agent.prompts import DEOBFUSCATION_SYSTEM_PROMPT, OUTPUT_SCHEMA
from kong.ghidra.types import BinaryInfo, StringEntry
from kong.llm.tools import DEOBFUSCATION_TOOLS, ToolExecutor
if TYPE_CHECKING:
+ from kong.agent.analyzer import AnalysisContext, LLMClient, LLMResponse
from kong.agent.queue import WorkItem
from kong.ghidra.client import GhidraClient
diff --git a/kong/evals/harness.py b/kong/evals/harness.py
index 1b21030..b014501 100644
--- a/kong/evals/harness.py
+++ b/kong/evals/harness.py
@@ -7,7 +7,7 @@ import re
from dataclasses import dataclass, field
from pathlib import Path
-from kong.evals.metrics import symbol_accuracy, type_accuracy
+from kong.evals.metrics import match_functions, symbol_accuracy, type_accuracy
@dataclass
@@ -58,40 +58,6 @@ def load_analysis(analysis_path: Path) -> list[dict[str, str]]:
return data["functions"]
-def match_functions(
- predicted: list[dict[str, str]],
- truth: list[dict[str, str]],
-) -> list[tuple[dict[str, str], dict[str, str] | None, float]]:
- """Match predicted functions to truth entries by best symbol_accuracy (greedy, no reuse)."""
- used_truth_indices: set[int] = set()
- results: list[tuple[dict[str, str], dict[str, str] | None, float]] = []
-
- scored_pairs: list[tuple[int, int, float]] = []
- for pi, pred in enumerate(predicted):
- for ti, tr in enumerate(truth):
- sc = symbol_accuracy(pred["name"], tr["name"])
- scored_pairs.append((pi, ti, sc))
-
- scored_pairs.sort(key=lambda x: x[2], reverse=True)
-
- matched_pred: dict[int, tuple[int, float]] = {}
- for pi, ti, sc in scored_pairs:
- if pi in matched_pred or ti in used_truth_indices:
- continue
- if sc > 0.0:
- matched_pred[pi] = (ti, sc)
- used_truth_indices.add(ti)
-
- for pi, pred in enumerate(predicted):
- if pi in matched_pred:
- ti, sc = matched_pred[pi]
- results.append((pred, truth[ti], sc))
- else:
- results.append((pred, None, 0.0))
-
- return results
-
-
def score(analysis_path: Path, source_path: Path) -> Scorecard:
"""Load analysis and source, match functions, compute scores."""
raw = json.loads(analysis_path.read_text())
diff --git a/kong/evals/metrics.py b/kong/evals/metrics.py
index 70c02d1..dacc9f3 100644
--- a/kong/evals/metrics.py
+++ b/kong/evals/metrics.py
@@ -4,8 +4,6 @@ from __future__ import annotations
import re
-from kong.evals.harness import match_functions
-
_SYNONYM_GROUPS: list[set[str]] = [
{"search", "find", "lookup"},
@@ -186,6 +184,40 @@ def type_accuracy(predicted_sig: str, truth_sig: str) -> float:
return score
+def match_functions(
+ predicted: list[dict[str, str]],
+ truth: list[dict[str, str]],
+) -> list[tuple[dict[str, str], dict[str, str] | None, float]]:
+ """Match predicted functions to truth entries by best symbol_accuracy (greedy, no reuse)."""
+ used_truth_indices: set[int] = set()
+ results: list[tuple[dict[str, str], dict[str, str] | None, float]] = []
+
+ scored_pairs: list[tuple[int, int, float]] = []
+ for pi, pred in enumerate(predicted):
+ for ti, tr in enumerate(truth):
+ sc = symbol_accuracy(pred["name"], tr["name"])
+ scored_pairs.append((pi, ti, sc))
+
+ scored_pairs.sort(key=lambda x: x[2], reverse=True)
+
+ matched_pred: dict[int, tuple[int, float]] = {}
+ for pi, ti, sc in scored_pairs:
+ if pi in matched_pred or ti in used_truth_indices:
+ continue
+ if sc > 0.0:
+ matched_pred[pi] = (ti, sc)
+ used_truth_indices.add(ti)
+
+ for pi, pred in enumerate(predicted):
+ if pi in matched_pred:
+ ti, sc = matched_pred[pi]
+ results.append((pred, truth[ti], sc))
+ else:
+ results.append((pred, None, 0.0))
+
+ return results
+
+
def overall_score(
predicted: list[dict[str, str]],
truth: list[dict[str, str]],