find-CPhysBox_Use_PatchCaller: Скилл поиска и патчинга

--- name: find-CPhysBox_Use_PatchCaller description: | Find and identify the CPhysBox_Use_PatchCaller patch in CS2 binary using IDA Pro MCP. This patch changes the third argument of the sub_XXXXXXXX call inside CPhysBox_Use from inputdata->pCaller to this (the CPhysBox entity pointer), so the caller is the physbox itself. Use this skill when reverse engineering CS2 server.dll or libserver.so to locate and patch the caller argument. Trigger: CPhysBox_Use patch caller, CPhysBox_Use_PatchCaller, physbox use caller patch disable-model-invocation: true --- # CPhysBox_Use_PatchCaller Patch Workflow ## Overview Locate the instruction inside `CPhysBox_Use` that loads `inputdata->pCaller` (the third argument to the inner call) and generate a patch that replaces it with `this` (the CPhysBox entity pointer). The target code pattern in pseudocode (before patch): ```c __int64 __fastcall CPhysBox_Use(__int64 pthis, __int64 inputdata) { __int64 result; result = CBaseEntity_Use(pthis, inputdata); if ( *(_BYTE *)(pthis + 2093) ) return sub_XXXXXXXX((int)pthis + 2176, *(_QWORD *)inputdata, *(_QWORD *)(inputdata + 8)); // ^^^ patch target: pCaller return result; } ``` After patch: ```c return sub_XXXXXXXX((int)pthis + 2176, *(_QWORD *)inputdata, pthis); // ^^^ now uses this ``` ### Windows The target instruction: ```asm mov r8, [rbx+8] ; r8 = inputdata->pCaller ``` Patched to: ```asm mov r8, rdi ; r8 = this (pthis) nop ``` - Original bytes: `4C 8B 43 08` - Patch bytes: `49 89 F8 90` ### Linux The target instruction: ```asm mov rdx, [r12+8] ; rdx = inputdata->pCaller ``` Patched to: ```asm mov rdx, rbx ; rdx = this (pthis) nop nop ``` - Original bytes: `49 8B 54 24 08` - Patch bytes: `48 89 DA 90 90` ## Prerequisites - `CPhysBox_Use` must already be identified. Use SKILL `/get-func-from-yaml` with `func_name=CPhysBox_Use` to load its address. If YAML does not exist, run SKILL `/find-CPhysBox_Use` first. ## Location Steps ### 1. Get CPhysBox_Use Function Address **ALWAYS** Use SKILL `/get-func-from-yaml` with `func_name=CPhysBox_Use`. If the skill returns an error, stop and report to user. ### 2. Decompile and Locate the Caller Argument Decompile the function: ``` mcp__ida-pro-mcp__decompile(addr="<func_va>") ``` In the decompiled output, search for the inner call pattern. The key indicators are: - A call to `CBaseEntity_Use(pthis, inputdata)` at the beginning - A byte check: `if ( *(_BYTE *)(pthis + <offset>) )` - Inside the if-block: a call with three arguments where the third is `*(_QWORD *)(inputdata + 8)` — this is `inputdata->pCaller` Note the address annotation on the line that loads `inputdata + 8`. ### 3. Disassemble Around the Caller Load Disassemble the function to find the exact instruction that loads `inputdata->pCaller`: ``` mcp__ida-pro-mcp__disasm(addr="<func_va>", max_instructions=30) ``` **Windows pattern:** ```asm mov r8, [rbx+8] ; load inputdata->pCaller into r8 (3rd arg) lea rcx, [rdi+880h] ; 1st arg mov rdx, [rbx] ; 2nd arg = inputdata->pActivator call sub_XXXXXXXX ``` **Linux pattern:** ```asm mov rdx, [r12+8] ; load inputdata->pCaller into rdx (3rd arg in SysV ABI) ... call sub_XXXXXXXX ``` Record: - **patch_va**: Address of the `mov r8, [rbx+8]` (Windows) or `mov rdx, [r12+8]` (Linux) instruction ### 4. Determine Patch Bytes Read the original bytes of the instruction: ``` mcp__ida-pro-mcp__get_bytes(regions={"addr": "<patch_va>", "size": 5}) ``` **Windows** (`mov r8, [rbx+8]` — 4 bytes: `4C 8B 43 08`): - Patch bytes: `49 89 F8 90` (`mov r8, rdi` + `nop`) **Linux** (`mov rdx, [r12+8]` — 5 bytes: `49 8B 54 24 08`): - Patch bytes: `48 89 DA 90 90` (`mov rdx, rbx` + 2x `nop`) ### 5. Generate Patch Signature **ALWAYS** Use SKILL `/generate-signature-for-patch` to generate and validate the signature. Required context for the skill: - `func_name`: `CPhysBox_Use` - `func_va`: From step 1 - `patch_va`: Address of the mov instruction from step 3 - `original_instruction`: `mov r8, [rbx+8]` (Windows) or `mov rdx, [r12+8]` (Linux) - `patched_instruction`: `mov r8, rdi; nop` (Windows) or `mov rdx, rbx; nop; nop` (Linux) - `description`: `Patch CPhysBox_Use to pass this instead of inputdata->pCaller as the third argument` ### 6. Write YAML Output **ALWAYS** Use SKILL `/write-patch-as-yaml` to persist the results. Required parameters: - `patch_name`: `CPhysBox_Use_PatchCaller` - `patch_sig`: The validated signature from step 5 - `patch_bytes`: The computed patch bytes from step 4 - `patch_sig_disp`: From step 5 result (omit if 0) ## Output YAML Files - `CPhysBox_Use_PatchCaller.windows.yaml` - `CPhysBox_Use_PatchCaller.linux.yaml`