Take a ScarletNX crash report, svcBreak or hang from the log to the mechanism that causes it, using the symbolizer and disassembler on the game's code. Use when the game crashes (signal 11 or 7), calls svcBreak or fatal, hangs, or when a log shows "*** Crash". Also for corrupted data showing up in game structures.
中文简介正在准备中。当前显示原文说明。
实测评分 未实测
查看安装方法安装前可以查看交给 AI 代理的指令内容。
The goal is the mechanism: which instruction read what, where the bad value came from, and what the runtime did that real hardware wouldn't. A fix without that is a guess.
*** Crash (signal 11) at host pc=0x7f..., access to 0x7f... (game address cf8617f8) ***
inside the recompiled code of game function cf195f0 (+41b on the host)
thread 131 ()
x0 = ... x30 = ...
pc=cf34380 lr=cf1aacc sp=... fp=...
pc <NvDdkVicExecuteSync+320> lr <NvRmStreamFlushSync+1c>
#0 cf1aacc <NvRmStreamFlushSync+1c>
code ADDR: words...
mem xN ADDR: what the register points at
pc is the last call site the runtime recorded, often in the
caller, so start from the game function.in the runtime itself (exe offset +OFF). The presets build with symbols:
addr2line -e build/scarletnx -f -C -i 0xOFF (the build that was deployed).svcBreak means the SDK's own assert failed. The interesting code is the caller's check, just before the
svc #0x26.tools/symbols.py work/<Game> ADDR... # the crash report's game addresses
tools/disasm.py work/<Game> NvRmStreamFlushSync # a whole function
tools/disasm.py work/<Game> START END # a range
Both read the modules through the recompiler's own loader (build/scarletnx --dump-modules, or the binary
SNX_SCARLETNX names), so the build must exist. The dump is cached per game folder in ~/.cache/scarletnx/module-dumps.
Find the faulting instruction (the load or store whose address is the game address) and walk back to where its base register came from. Every value is either loaded from memory (from where, written by whom), passed in, or computed. Keep going until you reach something the runtime provided: an IPC answer, an ioctl result, shared memory the runtime writes, a buffer a device filled.
Large functions: disassemble only the part you need. Use the crash-triage agent to do the walk in its own context
when it takes many steps, and the sdk-scout agent to work out what an SDK function expects.
When it reproduces, look at it live. Start the game with -S (or -s, to stop before the runtime starts), then use
the scarletnx MCP server's GDB tools: gdb_attach with the game, gdb_threads (every host thread, and each guest
thread's state and guest pc), gdb_guest (a guest thread's registers and guest backtrace) and gdb_command
(thread apply all bt shows which lock each runtime thread waits on, for a hang). A crash stops at crash_report.
Add --gdb_jit=1 when the frames are in code the JIT compiled, at the cost of a stop per compile. The stub takes one
connection per run.
| Shape | Where to look |
|---|---|
A pointer holds floats (3f800000... is 1.0) or other data | Memory reused or written by something else: an out-of-bounds write by the game because the runtime gave it an out-of-range value (the 128-entry syncpoint table), a GPU write-back (render target, compute, DMA) landing on memory the game reused, or a race between game threads |
svcBreak right after a service call | The service answered something real hardware never does. Read the assert's condition |
A wait that never ends (threads.txt) | An event the runtime never signals, or a release the game waits for (audio buffers) |
| Works once, crashes on the second movie, scene or submit | State the runtime keeps between uses: fixed ids, unreset buffers |
Access to (nil) with a game address near ffff8... | A guest null or garbage pointer translated: look at the register, not the host address |
Write down the chain from the runtime's behavior to the faulting instruction, with the addresses you read. Then fix
the runtime's side so it behaves like the console. Verify with repeat runs (snx-run), with the crash rate before and
after in the commit message. If the chain has a link you inferred rather than read, say so in the commit.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Audits uncommitted changes, branches, or PRs for bugs, edge cases, SOLID/DRY/SRP, architecture, security, performance, and CLAUDE.md violations — then automatically fixes ALL issues and re-audits in a loop until clean. Use when: audit, review, code review, review this PR, check code, look over this, inspect code, or any variation.
中文简介正在准备中。当前显示原文说明。
Record what ScarletNX work taught, where the next session will look for it. Use after a bug took more than one attempt to find, when the same mistake happened twice, when a debugging technique worked, or when the user corrects how work is done.
中文简介正在准备中。当前显示原文说明。
Build the ScarletNX runtime and recompile games correctly, avoiding the setups that build "successfully" into something that doesn't work. Covers the CPM.cmake dependencies, the Clang-only presets, the deploy into games/, when a game needs recompiling, and the warnings a plain build doesn't show. Use when building, rebuilding, recompiling a game, after pulling or bumping a dependency, or when a build result looks suspicious.
中文简介正在准备中。当前显示原文说明。
RFC #142's C rules: strict ISO C99, no redundant state, indices not pointer chains, one allocation per owner, atomics apart, no stored pointers, one global struct per file. Use when writing or reviewing C in src/.
中文简介正在准备中。当前显示原文说明。
Write and optimize ScarletNX's own compute shaders (astc.comp, vic.comp, future ones) for what the GPU actually executes: constant tables, bit twiddling, uniform branches, fewer fetches, workgroup and subgroup structure, exactness proofs, and the GLSL-as-C++ tests and lavapipe timing that check them. Use when writing, reviewing or speeding up a GLSL shader under src/nx/gpu/, or when deciding whether to split one.
中文简介正在准备中。当前显示原文说明。
Run a recompiled game unattended and judge the result: scripted button input, timed exit, screenshots, reading the log, and repeat runs for intermittent crashes. Use when verifying a runtime change, reproducing a bug, checking a scene renders, measuring a crash rate, or when the user asks to run or test the game.
中文简介正在准备中。当前显示原文说明。