Add num_api_calls heuristic

[kunalsz]

In reference to the discussion here #1212

What it is?

extract_num_api_calls is a function-level heuristic that counts how many API calls a function makes

• Decoding/decryption routines usually do a lot of tight arithmetic / bitwise operations over buffers(like xor)
• They often operate mostly on memory and registers and may call very few external/library APIs (sometimes none)
• In contrast, “normal” higher level functions (UI, file I/O, networking) often make many API/library calls

Low API call count : more consistent with a decoding rpoutine
High API call count : less consistent with decoding

How is it implemented ?

• count call instructions where vivisect resolved the target AND the target is not within the current function and not within the same binary’s code (it’s an import / external)
• also count calls “to” imported symbols by checking for a name like kernel32.CreateFileW or similar

Test results

With the heuristic in place the score has improved(same num of strings found)
Tests were executed on tests/data/malware/stackstrings/a294620543334a721a2ae8eaaf9680a0786f4b9a216d75b55cfd28f39e9430ea.exe_

• Without this heuristic:

"functions": {
            "analyzed_decoded_strings": 10,
            "analyzed_stack_strings": 8,
            "analyzed_tight_strings": 2,
            "decoding_function_scores": {
                "4198400": {
                    "score": 0.543,
                    "xrefs_to": 0
                },
                "4198864": {
                    "score": 0.554,
                    "xrefs_to": 1
                },
                "4199600": {
                    "score": 0.764,
                    "xrefs_to": 1
                },
                "4199696": {
                    "score": 0.456,
                    "xrefs_to": 1
                },
                "4199728": {
                    "score": 0.95,
                    "xrefs_to": 25
                },
                "4200272": {
                    "score": 0.472,
                    "xrefs_to": 2
                },
                "4200976": {
                    "score": 0.472,
                    "xrefs_to": 2
                },
                "4203168": {
                    "score": 0.634,
                    "xrefs_to": 3
                },
                "4205184": {
                    "score": 0.751,
                    "xrefs_to": 6
                },
                "4205328": {
                    "score": 0.804,
                    "xrefs_to": 3
                }
            }

"runtime": {
            "decoded_strings": 3.8492,
            "find_features": 0.1299,
            "language_strings": 0,
            "stack_strings": 0.2986,
            "start_date": "2026-02-13T13:20:57.460886Z",
            "static_strings": 0.0004,
            "tight_strings": 0.2559,
            "total": 7.9169,
            "vivisect": 3.3805
        }

With this heuristic :

"functions": {
            "analyzed_decoded_strings": 10,
            "analyzed_stack_strings": 8,
            "analyzed_tight_strings": 2,
            "decoding_function_scores": {
                "4198400": {
                    "score": 0.6,
                    "xrefs_to": 0
                },
                "4198864": {
                    "score": 0.61,
                    "xrefs_to": 1
                },
                "4199600": {
                    "score": 0.788,
                    "xrefs_to": 1
                },
                "4199696": {
                    "score": 0.547,
                    "xrefs_to": 1
                },
                "4199728": {
                    "score": 0.953,
                    "xrefs_to": 25
                },
                "4200272": {
                    "score": 0.56,
                    "xrefs_to": 2
                },
                "4200976": {
                    "score": 0.56,
                    "xrefs_to": 2
                },
                "4203168": {
                    "score": 0.68,
                    "xrefs_to": 3
                },
                "4205184": {
                    "score": 0.762,
                    "xrefs_to": 6
                },
                "4205328": {
                    "score": 0.822,
                    "xrefs_to": 3
                }
            },
            "discovered": 10,
            "library": 0
        }
    }

"runtime": {
            "decoded_strings": 3.9446,
            "find_features": 0.0661,
            "language_strings": 0,
            "stack_strings": 0.3221,
            "start_date": "2026-02-13T13:36:02.831242Z",
            "static_strings": 0.0004,
            "tight_strings": 0.3008,
            "total": 8.1679,
            "vivisect": 3.5308
        }

@mr-tz Looking forward to your insights !

[gemini-code-assist]

Summary of Changes

Hello @kunalsz, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request enhances the FLOSS tool's capability to detect string decoding functions by incorporating a novel heuristic based on the number of external API calls made by a function. This addition aims to refine the scoring of functions, leveraging the observation that decoding routines typically perform low-level operations with minimal external API interaction, thereby improving the accuracy of identifying such functions.

Highlights

• New Heuristic: NumAPICalls: Introduced a new function-level heuristic, NumAPICalls, to improve the identification of string decoding routines by counting external API calls.
• API Call Counting Logic: Implemented helper functions (_iter_call_targets, _looks_like_import_name, _is_external_api_target) to accurately identify and count external API calls, distinguishing them from internal function calls.
• Scoring Mechanism: Defined a scoring mechanism for NumAPICalls where functions with fewer API calls receive a higher score, indicating a stronger likelihood of being a decoding routine.
• Integration: Integrated the extract_num_api_calls function into the FUNCTION_HANDLERS list, enabling its automatic application during feature extraction.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Changelog

• floss/features/extract.py
  • Added NumAPICalls to the imported features.
  • Implemented helper functions _iter_call_targets, _looks_like_import_name, and _is_external_api_target to identify external API call targets.
  • Developed extract_num_api_calls to count these external API calls within a function.
  • Included extract_num_api_calls in the FUNCTION_HANDLERS tuple.
• floss/features/features.py
  • Defined the NumAPICalls class, inheriting from Feature.
  • Assigned a LOW weight to the NumAPICalls feature.
  • Implemented the score method for NumAPICalls, providing a tiered scoring logic based on the count of API calls.

Activity

• The changes were initiated following a discussion in issue Add Complex Functions heuristic #1212.
• Insights and review were specifically requested from @mr-tz.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces a new heuristic, num_api_calls, to help identify string decoding functions based on the number of API calls they make. The implementation is sound, but I've identified a high-severity bug that would limit its effectiveness on non-x86 architectures, and a medium-severity issue related to error handling that could make debugging difficult. I have provided specific suggestions to address these points.

[gemini-code-assist]

The check for mnem == "call" is specific to x86/x64 architectures and will not work for other architectures like ARM, where call instructions are BL. This makes the heuristic less portable. The _iter_call_targets function already correctly identifies call-like instructions using the architecture-agnostic envi.BR_PROC flag, so this check is both redundant and buggy.

[gemini-code-assist]

Swallowing exceptions with a broad except Exception: and a return or pass can hide bugs and makes debugging difficult. It is better to log the exception to provide visibility into potential issues during analysis. This pattern is also present in _is_external_api_target.

Suggested change

	except Exception:
	return
	except Exception as e:
	logger.warning("Failed to get branches for instruction at 0x%x: %s", insn.va, e)
	return

[mr-tz]

Thanks for the proposal. On closer look, and seeing the already noticeable slowdown, let's hold off with additional heuristics unless these really provide a benefit in detection more detection functions. What do you think?

[kunalsz]

Thanks for the feedback! I understand the concern and just wanted to highlight that the time difference added is quite small and more like applying a filter , it improves the confidence in detection without significantly altering existing logic. I still believe it adds value, but I'm happy to adapt based on your suggestions!