[alfalfa-star]

What

Implements the four libSceAgcDriver capture exports that still called NotImplemented_nid_no_patch:

• sceAgcDriverRequestCaptureStart, sceAgcDriverRequestCaptureStop, sceAgcDriverTriggerCapture take no arguments and return DEBUG_UNAVAILABLE (0x8A6C1000), since no capture tool is attached. OpenAGC's agcdriver.h and SharpProspero's SceAgcDriver.cs both declare RequestCaptureStart without arguments. sceAgcDriverIsCaptureInProgress already returns false.
• sceAgcDriverGetShaderDebuggingStatus returns 1.

Sources: OpenAGC returns the same code for the capture calls; whether a console with no capture attached does the same is unconfirmed. The value 1 is the constant OpenAGC reads from firmware 5.50 (rJUyMrDdxJg at 0x6740); what it means is unknown. Both are recorded in TechnicalDebt.

Tested

Linux (Arch, GCC), Release build, on main 57f96fd:

• guest_agc_driver_resources: checks the three capture calls' error code, that no capture is reported in progress afterwards, and the debugging status value. It aborts without this change (exit 134).
• Full ctest on main 90c152e merged with this branch, feat(agc): implement the occlusion query size queries #2848 and feat(npwebapi2): implement sceNpWebApi2SetRequestTimeout #2853: 813/813 passed (4 skipped for lack of a display).
• tools/check_conventions.py --base upstream/main passed.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed in TechnicalDebt
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none. Adds the same DEBUG_UNAVAILABLE constant to Status.cpp and touches the same test as feat(agc-driver): report submit validation as unavailable #2849, so whichever merges second needs a rebase
• AI-assisted: yes

🤖 Generated with Claude Code

[github-actions]

Progress report (66d577d → ae327c4)

📈 System libraries: 85.56% (+0.12%, +4 implemented functions)

✅ 4 implemented

Library	Name
libSceAgcDriver	sceAgcDriverGetShaderDebuggingStatus
libSceAgcDriver	sceAgcDriverRequestCaptureStart
libSceAgcDriver	sceAgcDriverRequestCaptureStop
libSceAgcDriver	sceAgcDriverTriggerCapture

[alfalfa-star]

The one failing job, write-watch (macos-15-intel), is the same write_watch_threads timeout as on #2699. The three repeats took 16.8 s, 23.3 s and then hit the 30 s limit. This PR only touches libSceAgcDriver and GuestAgcDriverResources.cpp, which that job does not build. In the last 100 Build runs this job failed 9 times, once on main. Could someone with rights rerun it?

[oneandonlydean]

Same as #2849: which title calls these? For sceAgcDriverGetShaderDebuggingStatus the title would also show what the 1 means to it.

[screamm]

Windows validation of the current head 69be6331 merged onto main 91b13413.

Windows 11 Pro 26220, WinLibs MinGW-w64 GCC 15.2.0 r7, Ninja, Release, BUILD_TESTING=ON, full build including the libs target, NVIDIA RTX 4070.

• ctest -j8 -E "^agc_driver_|^video_out_flip": 478/479, including guest_agc_driver_resources. The one failure was the GPU test agc_driver with a heap-corruption exit under -j8 load; it passes 2/2 when run alone on the same binaries, so I count it as load-related and unrelated to this change.
• The other GPU execution tests (agc_driver_*) were not run; two of them fail on unchanged main on this host for unrelated reasons.

Validation run with Claude Code assistance.

[adriadam10]

Thanks for the Windows run and the clear TechnicalDebt entry. Before this can go in:

• No title is named. @oneandonlydean asked which title calls these exports (and, for sceAgcDriverGetShaderDebuggingStatus, what the 1 means to it) and that is still unanswered. Please add the title ID and the log line where it stops on main.
• core/libs/prx/libSceAgcDriver/State/src/Status.cpp:26,30,34,38: the returned values (0x8A6C1000 for the capture calls, 1 for the debugging status) come from one source (OpenAGC) and are unconfirmed on a console, as the PR itself says. Unverified codes should keep throwing until a title or a second source confirms them.
• Status.cpp:24: sceAgcDriverRequestCaptureStart takes const char* here, but SharpProspero declares it without arguments (Interop/Agc/SceAgcDriver.cs:206). Please say which one the signature follows and why.

The red write-watch (macos-15-intel) job is the known write_watch_threads flake, not this change. Happy to look again once a title and a confirmed value are in.

[alfalfa-star]

No title I can run reaches the capture exports on current main (90c152e):

• LEGO Horizon Adventures (PPSA14632) aborts after about 35 s at color views of several array slices are unsupported (Layered rendering: a draw into all slices of a 3D color target aborts with "color views of several array slices are unsupported" #2102). My local test branch gets past that and other stops; there it played 385 s with scripted input, through New Game and a save, without calling the capture exports.
• Kena: Bridge of Spirits (PPSA01802) crashes in game code after about 60 s, with no not implemented line before it.

I'm converting this to a draft until a title stops at it. Then I'll post the title ID and the log line and mark it ready.

Changed:

• sceAgcDriverRequestCaptureStart now takes no arguments. OpenAGC's agcdriver.h and SharpProspero's SceAgcDriver.cs both declare it that way; the const char* came from main's earlier stub. The test calls it without arguments.
• The entry moved under Silent stubs, as main's conventions check now requires.

On the values: I found no second source. sharpemu, REmu and shadPS4 only list the names, and prosper has no golden for them. They stay as in the PR until a title reaching them shows how it treats the result.

Tested: Linux, GCC, Release, RX 9070 XT (RADV): full ctest 813/813 (4 video_out_flip_* skipped) on main 90c152e merged with this branch, #2848 and #2853; tools/check_conventions.py passes.

AI-assisted: yes (Claude Code)

[adriadam10]

Thanks for the signature fix and the TechnicalDebt entry; both look right, and guest_agc_driver_resources fails without the change and passes with it. Two points are still open:

• No title is named. @oneandonlydean's question (which title calls sceAgcDriverRequestCaptureStart/Stop/TriggerCapture and sceAgcDriverGetShaderDebuggingStatus, and what the 1 means to it) is unanswered. Please add the title ID and the call site or log line where it stops on main; Astro's Playroom does not import any of them, so we cannot check it from our side.
• core/libs/prx/libSceAgcDriver/State/src/Status.cpp:24,28,32,36: 0x8A6C1000 and the constant 1 come from a single source (OpenAGC) and are unconfirmed on a console. Where main threw, returning a plausible unverified value is not exact-or-throw. Either a second source or a title's observed behaviour, or keep these throwing.

[oneandonlydean]

Same as #2849: closing until a title reaches these; the title will also settle the return values.

You're welcome to reopen this if you think it was closed by mistake or the situation changes.