Use when you have a written implementation plan to execute in a separate session with review checkpoints
npx skills add hzqst/CS2_VibeSignatures --skill "generate-signature-for-function"
Install specific skill from multi-skill repository
# Description
|
# SKILL.md
name: generate-signature-for-function
description: |
Generate and validate unique byte signatures for functions using IDA Pro MCP. Use this skill when you need to create a pattern-scanning signature for a function that can reliably locate it across binary updates.
Triggers: generate signature, byte signature, pattern signature, function signature, unique signature, sig for function
Generate Signature for Function
Generate a unique hex byte signature for a function that can be used for pattern scanning.
Prerequisites
- Function address (from decompilation, xrefs, or rename)
- IDA Pro MCP connection
Method
1. Get Function Bytes and Disassembly
First, retrieve the raw bytes and disassembly to understand the function structure:
mcp__ida-pro-mcp__py_eval code="""
import ida_bytes
func_addr = <func_addr>
# Get first 64 bytes of function
raw_bytes = ida_bytes.get_bytes(func_addr, 64)
print("Function bytes:", ' '.join(f'{b:02X}' for b in raw_bytes))
"""
Also get disassembly for context:
mcp__ida-pro-mcp__disasm addr="<func_addr>" max_instructions=15
2. Analyze and Generate Signature (LLM Task)
YOU (the LLM) must analyze the bytes and disassembly to create a signature.
When analyzing, consider:
- Opcodes (instruction bytes): Usually stable, keep as-is
- Immediates/Offsets: Often change between builds, use ?? wildcards
- Register encodings: Usually stable unless compiler changes register allocation
- Relocation addresses: Always use ?? wildcards (4 bytes for 32-bit, 8 for 64-bit)
- Displacement values in memory operands: May change, consider wildcarding, ESPECIALLY with E8 call, E9 jmp
Example analysis:
Bytes: 48 89 5C 24 08 48 89 74 24 10 57 48 83 EC 20 48 8B F9 E8 12 34 56 00
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^^^
Prologue - stable opcodes Call with relative offset - wildcard it
Result: 48 89 5C 24 08 48 89 74 24 10 57 48 83 EC 20 48 8B F9 E8 ?? ?? ?? ??
Generate a signature that is:
- Long enough to be unique (typically 16-32 bytes)
- Uses ?? for bytes that may change between binary updates
- Starts from function entry point
3. Validate Signature Uniqueness
Test that YOUR generated signature matches ONLY this function:
mcp__ida-pro-mcp__py_eval code="""
import idaapi
import ida_bytes
import ida_segment
func_addr = <func_addr>
# YOUR GENERATED SIGNATURE HERE (space-separated hex with ?? for wildcards)
signature_str = "<YOUR_SIGNATURE>" # e.g., "48 89 5C 24 08 48 89 74 24 10 57 48 83 EC 20 48 8B F9 E8 ?? ?? ?? ??"
# IDA pattern syntax uses single '?' wildcard tokens (not '??').
pattern_str = " ".join("?" if p == "??" else p for p in signature_str.split())
# Get .text segment bounds
seg = ida_segment.get_segm_by_name(".text")
start = seg.start_ea
end = seg.end_ea
# Compile pattern and search using IDA's native binary search (fast, avoids get_bytes() timeouts)
pat = ida_bytes.compiled_binpat_vec_t()
ida_bytes.parse_binpat_str(pat, start, pattern_str, 16)
flags = ida_bytes.BIN_SEARCH_FORWARD | ida_bytes.BIN_SEARCH_NOBREAK
matches = []
res = ida_bytes.bin_search3(start, end, pat, flags)
ea = res[0] if isinstance(res, tuple) else res # IDA 9 may return (ea, idx)
while ea != idaapi.BADADDR:
matches.append(ea)
res = ida_bytes.bin_search3(ea + 1, end, pat, flags)
ea = res[0] if isinstance(res, tuple) else res
print(f"Signature matches: {len(matches)}")
for m in matches:
print(hex(m))
if len(matches) == 1 and matches[0] == func_addr:
print("SUCCESS: Signature is unique and matches target function!")
elif len(matches) == 1 and matches[0] != func_addr:
print("WARNING: Signature matches but at different address, This should not happen!")
elif len(matches) > 1:
print("FAILED: Signature not unique, need longer/different pattern. Ignore this failure if we got SUCCESS before, as py_eval has issue with duplicate evaluation sometimes.")
elif len(matches) == 0:
print("FAILED: Found nothing with this signature. Ignore this failure if we got SUCCESS before, as py_eval has issue with duplicate evaluation sometimes.")
"""
4. Iterate if Needed
If the signature is not unique:
1. Extend the signature length
2. Analyze which bytes are causing false matches
3. Add more specific bytes or adjust wildcard placement
4. Re-validate until unique
5. Continue with Unfinished Tasks
If we are called by a task from a task list / parent SKILL, restore and continue with the unfinished tasks.
Output Format
Signature format: space-separated hex bytes with ?? for wildcards.
Example: 48 89 5C 24 08 48 89 74 24 10 57 48 83 EC ?? 48 8B F9 E8 ?? ?? ?? ??
Signature Analysis Guidelines for LLM
When analyzing function bytes, consider these instruction patterns:
Bytes to Keep (Usually Stable)
- Function prologues:
48 89 5C 24(mov [rsp+x], rbx),55(push rbp),48 83 EC(sub rsp) - Opcode bytes: The actual instruction opcodes
- ModR/M bytes: Register-to-register operations
- Fixed immediate values: Constants that are part of the algorithm
Bytes to Wildcard (May Change)
- Relative call/jump offsets: 4 bytes after
E8(call) orE9(jmp) - these are relative addresses - Absolute addresses: Any pointer/address in code
- Stack offsets in large functions: May change with local variable layout
- String/data references: Offsets to .rdata/.data sections
Common Patterns
| Pattern | Meaning | Wildcard? |
|---|---|---|
E8 XX XX XX XX |
Relative call | Wildcard the 4 offset bytes |
48 8D 0D XX XX XX XX |
LEA with RIP-relative | Wildcard the 4 offset bytes |
FF 15 XX XX XX XX |
Indirect call via IAT | Wildcard the 4 offset bytes |
48 8B 05 XX XX XX XX |
MOV from global | Wildcard the 4 offset bytes |
Important
DO NOT use find_bytes to validate signatures - it doesn't work reliably for function pattern matching. Always use the py_eval method with mask support above.
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.