binary-triage（二元分类）

Binary Triage

Instructions

We are triaging a binary to quickly understand what it does. This is an initial survey, not deep analysis. Our goal is to:

1. Identify key components and behaviors
2. Flag suspicious or interesting areas
3. Create a task list of next steps for deeper investigation

Binary triage with ReVa

Follow this systematic workflow using ReVa's MCP tools:

1. Identify the Program

• Use get-current-program to see the active program
• Or use list-project-files to see available programs in the project
• Note the programPath (e.g., "/Hatchery.exe") for use in subsequent tools

2. Survey Memory Layout

• Use get-memory-blocks to understand the binary structure
• Examine key sections:

- .text - executable code - .data - initialized data - .rodata - read-only data (strings, constants) - .bss - uninitialized data

• Flag unusual characteristics:

- Unusually large sections - Packed/encrypted sections - Executable data sections - Writable code sections

3. Survey Strings

• Use get-strings-count to see total string count
• Use get-strings with pagination (100-200 strings at a time)
• Look for indicators of functionality or malicious behavior:

- Network: URLs, IP addresses, domain names, API endpoints - File System: File paths, registry keys, configuration files - APIs: Function names, library references - Messages: Error messages, debug strings, log messages - Suspicious Keywords: admin, password, credential, token, crypto, encrypt, decrypt, download, execute, inject, shellcode, payload

4. Survey Symbols and Imports

• Use get-symbols-count with includeExternal=true to count imports
• Use get-symbols with includeExternal=true and filterDefaultNames=true
• Focus on external symbols (imports from libraries)
• Flag interesting/suspicious imports by category:

- Network APIs: connect, send, recv, WSAStartup, getaddrinfo, curl_*, socket - File I/O: CreateFile, WriteFile, ReadFile, fopen, fwrite, fread - Process Manipulation: CreateProcess, exec, fork, system, WinExec, ShellExecute - Memory Operations: VirtualAlloc, VirtualProtect, mmap, mprotect - Crypto: CryptEncrypt, CryptDecrypt, EVP_*, AES_*, bcrypt, RC4 - Anti-Analysis: IsDebuggerPresent, CheckRemoteDebuggerPresent, ptrace - Registry: RegOpenKey, RegSetValue, RegQueryValue

• Note the ratio of imports to total symbols (heavy import usage may indicate reliance on libraries)

5. Survey Functions

• Use get-function-count with filterDefaultNames=true to count named functions
• Use get-function-count with filterDefaultNames=false to count all functions
• Calculate ratio of named vs unnamed functions (high unnamed ratio = stripped binary)
• Use get-functions with filterDefaultNames=true to list named functions
• Identify key functions:

- Entry points: entry, start, _start - Main functions: main, WinMain, DllMain, _main - Suspicious names: If not stripped, look for revealing function names

6. Cross-Reference Analysis for Key Findings

• For interesting strings found in Step 3:

- Use find-cross-references with direction="to" and includeContext=true - Identify which functions reference suspicious strings

• For suspicious imports found in Step 4:

- Use find-cross-references with direction="to" and includeContext=true - Identify which functions call suspicious APIs

• This helps prioritize which functions need detailed examination

7. Selective Initial Decompilation

• Use get-decompilation on entry point or main function

- Set limit=30 to get ~30 lines initially - Set includeIncomingReferences=true to see callers - Set includeReferenceContext=true for context snippets

• Use get-decompilation on 1-2 suspicious functions identified in Step 6

- Set limit=20-30 for quick overview

• Look for high-level patterns:

- Loops (encryption/decryption routines) - Network operations - File operations - Process creation - Suspicious control flow (obfuscation indicators)

• Do not do deep analysis yet - this is just to understand general behavior

8. Document Findings and Create Task List

• Use the TodoWrite tool to create an actionable task list with items like:

- "Investigate string 'http://malicious-c2.com' (referenced at 0x00401234)" - "Decompile function sub_401000 (calls VirtualAlloc + memcpy + CreateThread)" - "Analyze crypto usage in function encrypt_payload (uses CryptEncrypt)" - "Trace anti-debugging checks (IsDebuggerPresent at 0x00402000)" - "Examine packed section.UPX0 for unpacking routine"

• Each todo should be:

- Specific (include addresses, function names, strings) - Actionable (what needs to be investigated) - Prioritized (most suspicious first)

Output Format

Present triage findings to the user in this structured format:

Program Overview

• Name: [Program name from programPath]
• Type: [Executable type - PE, ELF, Mach-O, etc.]
• Platform: [Windows, Linux, macOS, etc.]

Memory Layout

• Total Size: [Size in bytes/KB/MB]
• Key Sections: [List main sections with sizes and permissions]
• Unusual Characteristics: [Any packed/encrypted/suspicious sections]

String Analysis

• Total Strings: [Count from get-strings-count]
• Notable Findings: [Bullet list of interesting strings with context]
• Suspicious Indicators: [URLs, IPs, suspicious keywords found]

Import Analysis

• Total Symbols: [Count from get-symbols-count]
• External Imports: [Count of external symbols]
• Key Libraries: [Main libraries imported]
• Suspicious APIs: [Categorized list of concerning imports]

Function Analysis

• Total Functions: [Count with filterDefaultNames=false]
• Named Functions: [Count with filterDefaultNames=true]
• Stripped Status: [Yes/No based on ratio]
• Entry Point: [Address and name]
• Main Function: [Address and name]
• Key Functions: [List of important functions identified]

Suspicious Indicators

[Bulleted list of red flags discovered, prioritized by severity]

Recommended Next Steps

[Present the task list created in Step 8]

• Each item should be specific and actionable
• Prioritize by severity/importance
• Include addresses, function names, and context

Important Notes

• Speed over depth: This is triage, not full analysis. Move quickly through steps.
• Use pagination: Don't request thousands of strings/functions at once. Use chunks of 100-200.
• Focus on anomalies: Flag things that are unusual, suspicious, or interesting.
• Context is key: When using cross-references, enable includeContext=true for code snippets.
• Create actionable todos: Each next step should be specific enough for another agent to execute.
• Be systematic: Follow all 8 steps in order for comprehensive coverage.