[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77QZC79GCCH3DNQSHJ8F4QB":3,"skill-same:01M77QZC79GCCH3DNQSHJ8F4QB":53},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":30,"commitSha":31,"skillDirectory":32,"forksCount":33,"repositoryHtmlUrl":34,"repositoryHomepageUrl":35,"ownerHtmlUrl":36,"ownerBlogUrl":6,"ownerDescription":37,"updatedAt":38,"install":39,"resources":52},"01M77QZC79GCCH3DNQSHJ8F4QB","peekaboo-linux-gnome-wayland-computer-use",null,"See and control GNOME Wayland desktops using gnome-screenshot, ydotool, Window Calls, AT-SPI hybrid element coordinates, and optional GNOME Remote Desktop. The Linux\u002FWayland counterpart to macOS computer-use (\"peekaboo\") — a native agent on the box can screenshot, reason over the image, locate UI elements by role\u002Fname, and click\u002Ftype them, with no human, no RDP, and no global-coordinate accessibility. Covers the native single-actor loop, the remote-over-SSH path, and an optional RDP fallback for hands-on human access.","Lord1Egypt\u002Fawesome-skill-forge","Lord1Egypt","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F99746359?v=4","01M7592YDNT0XK9WJ0T2MQVPDN","Mohamed Mounir Ali AL Keshita","01M75GRFNHC8VZNWEB8DKDE433",3,"MIT","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fcommunity\u002Fclawhub\u002Fp\u002Fpeekaboo-linux-gnome-wayland-computer-use\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: peekaboo-linux-gnome-wayland-computer-use\ndescription: See and control GNOME Wayland desktops using gnome-screenshot, ydotool, Window Calls, AT-SPI hybrid element coordinates, and optional GNOME Remote Desktop. The Linux\u002FWayland counterpart to macOS computer-use (\"peekaboo\") — a native agent on the box can screenshot, reason over the image, locate UI elements by role\u002Fname, and click\u002Ftype them, with no human, no RDP, and no global-coordinate accessibility. Covers the native single-actor loop, the remote-over-SSH path, and an optional RDP fallback for hands-on human access.\nmetadata: {\"clawdbot\":{\"emoji\":\"🖥️\",\"os\":[\"darwin\",\"linux\"]}}\n---\n\n# peekaboo-linux-gnome-wayland-computer-use\n\nThe Linux\u002FWayland counterpart to macOS computer-use — *\"see and poke the screen of a box for\ntroubleshooting or automation.\"* On macOS that's trivial (CGWindowList + CGEvent + AX tree). On\n**GNOME\u002FWayland it's locked down by design**, but it can be made to work end-to-end:\n\n- **See** → `gnome-screenshot` (re-enabled on GNOME 49+ via a small GNOME extension).\n- **Act** → `ydotool` (writes to `\u002Fdev\u002Fuinput`, no root).\n- **Target elements** → AT-SPI **WINDOW**-relative extents + the **Window Calls** extension's\n  window origin, composed into real absolute click coordinates (works where naive AT-SPI returns `(0,0)`).\n- **Hands-on (optional)** → GNOME Remote Desktop in Remote-Login mode for a human at an RDP client.\n\nBoth the \"see\" and \"act\" sides work **natively** (an agent process running ON the box) and\n**remotely** (from another machine over SSH). A native agent is a **single actor** — it can\nscreenshot, analyze, then ydotool-click in one loop. No second machine, no human, no RDP roundtrip.\n\nContext tags below — `[NATIVE]` = an agent process running ON the GNOME box (sharing the desktop\nsession of `user@1000`); `[REMOTE]` = on another machine, reaching the box over SSH.\n\n> Verified on Ubuntu 26.04 LTS + GNOME Shell 50 (Wayland), NVIDIA + AMD multi-GPU, multi-monitor.\n> Substitute your own host\u002Fuser where the examples use `USER@HOST` and `\u002Frun\u002Fuser\u002FUID`.\n\n## Capability matrix\n\n| Capability | Native (on box) | Remote (from another machine) |\n|---|---|---|\n| **Screenshot** | yes — `gnome-screenshot -f out.png` (extension unlocks it) | yes — same over SSH, then scp the file |\n| **Inject input** (mouse\u002Fkeys) | yes — ydotool via `\u002Fdev\u002Fuinput` (no root) | yes — over SSH, same ydotool |\n| **Click element by role\u002Fname** | yes — `locate-element` (AT-SPI WINDOW extents + Window Calls origin) | yes — same over SSH |\n| **Full see→reason→act loop** | yes — single actor, no RDP | yes — drive over SSH |\n| **Live interactive desktop** | n\u002Fa | yes — RDP client (optional, for a human) |\n\n## Target environment assumptions\n\n- A GNOME\u002FWayland box with a logged-in graphical session (the GUI session must be active on a seat).\n- For remote use: SSH access (key-based recommended), and the box's session env exported in\n  non-login shells — `XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID`,\n  `DBUS_SESSION_BUS_ADDRESS=unix:path=\u002Frun\u002Fuser\u002FUID\u002Fbus`, `WAYLAND_DISPLAY=wayland-0`, `DISPLAY=:0`.\n  (Replace `UID` with the session user's numeric id, e.g. `1000`.)\n- If agents run under `systemd --user` with the full graphical session env + lingering enabled,\n  they inherit all of the above and survive SSH disconnects.\n- `gnome-screenshot` grabs the **full virtual desktop** (all monitors stitched). On multi-4K\n  setups that's huge — use `-w` (focused window) or the per-display helper below.\n\n## CAPABILITY A — Screenshot (the key unlock)\n\nGNOME 49+ revoked `gnome-screenshot`'s access to the private Shell screenshot API, so on a stock\nbox every headless capture path is dead (see \"why\" below). The fix is the **`allow-gnome-screenshot`\nextension** (https:\u002F\u002Fgithub.com\u002Fsiddhpant\u002Fallow-gnome-screenshot, GNOME 49\u002F50): it re-adds\n`org.gnome.Screenshot` to the Shell's `_senderChecker._allowlistMap`, so the **`gnome-screenshot`\nbinary specifically** regains access. It is scoped to that one sender — it does NOT enable global\n`unsafe_mode`.\n\n`[NATIVE]` — just call it (session env already present):\n```bash\ngnome-screenshot -f \u002Ftmp\u002Fshot.png        # full virtual desktop (all monitors)\ngnome-screenshot -w -f \u002Ftmp\u002Fshot.png     # focused window only (smaller)\ngnome-screenshot -a -f \u002Ftmp\u002Fshot.png     # interactive area (needs a human at the screen — avoid headless)\n# then hand \u002Ftmp\u002Fshot.png to your image-analysis step\n```\n`[REMOTE]` — over SSH, then pull it back:\n```bash\nssh USER@HOST '\n  export XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID DBUS_SESSION_BUS_ADDRESS=unix:path=\u002Frun\u002Fuser\u002FUID\u002Fbus\n  gnome-screenshot -f \u002Ftmp\u002Fshot.png'\nscp USER@HOST:\u002Ftmp\u002Fshot.png \u002Ftmp\u002F   # then analyze it\n```\n\n**Sanity-check the capture** — a 4K full-desktop PNG is ~15-20 MB; a tiny\u002Fnear-empty file or an\nall-black image means the extension got disabled or the session dropped to the GDM greeter. Quick\nnon-black check:\n`python3 -c \"from PIL import Image; im=Image.open('\u002Ftmp\u002Fshot.png').convert('RGB'); d=list(im.getdata()); print(sum(1 for p in d[::997] if p!=(0,0,0)))\"` → should be well >0.\n\n### Installing the extension from scratch\n```bash\n# [NATIVE or over SSH with the session env exported]\nsudo apt-get install -y gnome-screenshot\nD=~\u002F.local\u002Fshare\u002Fgnome-shell\u002Fextensions\u002Fallow-gnome-screenshot@siddh.me\ngit clone --depth 1 https:\u002F\u002Fgithub.com\u002Fsiddhpant\u002Fallow-gnome-screenshot.git \u002Ftmp\u002Fags\nmkdir -p \"$D\" && cp -r \u002Ftmp\u002Fags\u002Fallow-gnome-screenshot@siddh.me\u002F* \"$D\u002F\"\nsudo systemctl restart gdm3      # reload the Shell so it discovers the extension (see WARNING below)\n# after the session returns:\nexport XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID DBUS_SESSION_BUS_ADDRESS=unix:path=\u002Frun\u002Fuser\u002FUID\u002Fbus\ngnome-extensions enable allow-gnome-screenshot@siddh.me\ngnome-extensions info allow-gnome-screenshot@siddh.me | grep State   # expect ACTIVE\n```\n\n### Per-display capture (capture ONE monitor, not the stitched canvas)\n\n`gnome-screenshot -f` grabs the **entire** virtual desktop. On multi-monitor setups that's a huge,\nmostly-empty canvas. To get a single clean display, use the bundled **`screenshot-display`** helper\n(`scripts\u002Fscreenshot-display.py`). It reads live monitor geometry from Mutter DisplayConfig and\ncrops the full capture to the chosen monitor.\n\n```bash\nscreenshot-display --list                               # enumerate monitors + geometry\nscreenshot-display --monitor primary --out \u002Ftmp\u002Fp.png   # primary (default)\nscreenshot-display --monitor DP-4   --out \u002Ftmp\u002Fm.png    # by connector name\nscreenshot-display --monitor 2      --out \u002Ftmp\u002Fm.png    # by index (0-based, sorted L->R, T->B)\nscreenshot-display --monitor active --out \u002Ftmp\u002Fw.png    # the FOCUSED WINDOW (best \"being worked on\" signal)\n```\nEach per-monitor crop is a clean single-display image — feed it straight to your vision step.\n`[REMOTE]`: run `screenshot-display` over SSH (export the session env), then scp the output back.\n\n### Window targeting — screenshot a specific program \u002F the display it's on\n\n`Shell.Eval`\u002F`Introspect` are locked, but the **Window Calls extension**\n(https:\u002F\u002Fgithub.com\u002Fickyicky\u002Fwindow-calls, GNOME 45-50) exposes window list + geometry over D-Bus\nat `\u002Forg\u002Fgnome\u002FShell\u002FExtensions\u002FWindows`. The `screenshot-display` helper uses it to find and crop\nto a window or its monitor:\n\n```bash\nscreenshot-display --list-windows                       # wm_class, id, focus, monitor, title\nscreenshot-display --window focused      --out \u002Ftmp\u002Fw.png   # crop to focused window's exact rect\nscreenshot-display --window \"wm_class=firefox\" --out \u002Ftmp\u002Fw.png  # match by wm_class substring\nscreenshot-display --window \"title=Settings\"   --out \u002Ftmp\u002Fw.png  # match by title substring\nscreenshot-display --window-display focused        --out \u002Ftmp\u002Fd.png  # the whole MONITOR that window sits on\nscreenshot-display --window-display \"wm_class=Code\" --out \u002Ftmp\u002Fd.png  # monitor a matched window is on\n```\nFlow: `List` → pick window (focused \u002F wm_class \u002F title) → `Details` gives `x,y,width,height,monitor`\n→ helper crops the full capture to the window rect, or to whichever monitor's rect contains the\nwindow's center. (`Details` works; `GetFrameBounds` is broken in this ext version — the helper\ndoesn't use it.)\n\n**Full setup\u002Fenablement guide** to replicate on any GNOME\u002FWayland box:\n`references\u002Fpeekaboo-and-rdp-setup-ubuntu-gnome-wayland.md`.\n\n## CAPABILITY B — Input injection (ydotool)\n\nydotool writes to `\u002Fdev\u002Fuinput` below the display server (the input analog of the screenshot\nunlock). Install it, run `ydotoold` as a persistent user service (socket\n`\u002Frun\u002Fuser\u002FUID\u002F.ydotool_socket`), and ACL `\u002Fdev\u002Fuinput` to your user (no root needed at runtime).\n\n`[NATIVE]`:\n```bash\nexport YDOTOOL_SOCKET=\u002Frun\u002Fuser\u002FUID\u002F.ydotool_socket\nydotool mousemove -a 960 540     # absolute pixel coords (no element targeting)\nydotool click 0xC0               # left click down+up\nydotool type \"hello\"\nydotool key 28:1 28:0            # Enter (keycode:press \u002F keycode:release)\n```\n`[REMOTE]`:\n```bash\nssh USER@HOST '\n  export XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID YDOTOOL_SOCKET=\u002Frun\u002Fuser\u002FUID\u002F.ydotool_socket\n  ydotool mousemove -a 700 400; ydotool click 0xC0; ydotool type \"hi\"'\n```\nIf `ydotoold.service` is down — `systemctl --user enable --now ydotoold.service`. For element\ntargeting (not blind coords) use the `locate-element` helper below.\n\n## CAPABILITY B+ — Click an element by role\u002Fname (SOM via hybrid coords)\n\n**You do NOT need blind pixel guessing or a vision round-trip for most clicks.** The bundled\n**`locate-element`** helper (`scripts\u002Flocate-element.py`) resolves a UI element to a real absolute\ncoordinate and clicks\u002Ftypes it.\n\n**Why it works when AT-SPI looks broken:** on Wayland, `getExtents(SCREEN)` returns `(0,0)` for\nGTK4\u002FElectron (no global coordinates by design), so naive AT-SPI clicking fails. But\n`getExtents(WINDOW)` (widget-relative) **is** populated, and **Window Calls** gives the window's\nscreen origin from Mutter's privileged view. The helper composes them:\n`abs = window_origin + window_relative_extents + size\u002F2`. Proven on GTK4 apps.\n\n```bash\nlocate-element --app gnome-text-editor --list                       # list elements + computed ABS coords\nlocate-element --app gnome-text-editor --role text --click          # click first text widget\nlocate-element --app org.gnome.Nautilus --name \"New Folder\" --click  # click by visible label\nlocate-element --app gnome-text-editor --role text --click --type \"hello\"   # click then type\nlocate-element --window-id 1367000216 --role \"push button\" --name Open --click  # by explicit window id\nlocate-element --app firefox --name Reload --print                  # just print \"X Y\", act yourself\n```\n- `--app` matches the AT-SPI app name AND wm_class loosely (normalizes `org.gnome.TextEditor` ↔ `gnome-text-editor`).\n- `--role`\u002F`--name` are case-insensitive substring filters; `--index N` picks the Nth match.\n- Needs `toolkit-accessibility=true` (`gsettings set org.gnome.desktop.interface toolkit-accessibility true`),\n  Window Calls active, and ydotoold running.\n\n**Vision fallback (only when an element zeros even WINDOW extents** — custom-drawn canvases, some\nElectron internals): `locate-element` exits non-zero and prints the fallback recipe —\n`screenshot-display --window 'wm_class=X' --out \u002Ftmp\u002Fw.png` → analyze for pixel coords →\n`ydotool mousemove -a -x X -y Y`.\n\nDeep dive on the technique, the gotchas, and why Newton isn't a substitute yet:\n`references\u002Fwayland-atspi-hybrid-som.md`.\n\n## The native loop (single actor — the whole point)\n\n```\n[NATIVE agent on the box]\n  1. gnome-screenshot -f \u002Ftmp\u002Fs.png\n  2. analyze \u002Ftmp\u002Fs.png  → \"the Save button is at ~(1820, 640)\"\n  3. ydotool mousemove -a 1820 640 && ydotool click 0xC0\n  4. gnome-screenshot again → verify it landed\n```\nNo second machine, no human, no RDP. For element targeting prefer `locate-element` over blind coords.\n\n## Window actions (focus \u002F move \u002F resize \u002F close)\n\nWindow Calls also does focus\u002Fmove\u002Fresize\u002Fclose verbs (window id from `--list-windows`):\n```bash\nWIN=\u002Forg\u002Fgnome\u002FShell\u002FExtensions\u002FWindows\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Activate \u003Cid>   # focus\u002Fraise\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Move \u003Cid> \u003Cx> \u003Cy>\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Resize \u003Cid> \u003Cw> \u003Ch>\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Maximize \u003Cid>\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Close \u003Cid>\n```\nFlow: `Activate \u003Cid>` → `screenshot-display --window focused` → `ydotool` interact.\n\n## CAPABILITY C — Live interactive desktop via RDP (optional, for a human)\n\nFor a person to drive the desktop hands-on from another machine, use **GNOME Remote Desktop** in\n**Remote Login** mode (system\u002Fheadless, fresh negotiated display — NOT Desktop Sharing, which\nmirrors the physical multi-monitor desktop and re-hits the huge-canvas problem).\n\n- Mode: `grdctl --system` (system service `gnome-remote-desktop.service`, enabled at boot).\n- Listener: `*:3389` — **always LAN-scope it in your firewall, never any-source.**\n- TLS cert at `\u002Fvar\u002Flib\u002Fgnome-remote-desktop\u002Frdp-tls.{crt,key}`.\n- Credentials: a device user + password over NLA. Setting the device cred = the Linux user's\n  password makes both stages (NLA device prompt, then GDM login) take the same credential.\n\nConnect from a client (Microsoft Remote Desktop \u002F Windows App, or any RDP client) → `HOST` → device\nuser\u002Fpass → GDM → sign in as the Linux user.\n\n**⚠️ #1 RDP gotcha — credential changes need a daemon restart.** `grdctl --system rdp set-credentials`\nwrites the keyfile but the **running daemon does NOT hot-reload it into its NTLM SAM**. Until you\n`sudo systemctl restart gnome-remote-desktop.service`, NLA fails for every client:\n`ntlm_fetch_ntlm_v2_hash: Could not find user in SAM database` → `SEC_E_NO_CREDENTIALS` →\n`transport_accept_nla: client authentication failure`. Meanwhile `grdctl status` shows the correct\ncreds (it reads the file). Clients report this as **`FREERDP_ERROR_CONNECT_TRANSPORT_FAILED`** (RoyalTS)\nor `ERRCONNECT_AUTHENTICATION_FAILED` (FreeRDP) — looks like transport\u002Fcert, is actually stale creds.\nFix: set creds (password via STDIN) → **restart the service** → verify journald shows no `SAM`\u002F\n`NO_CREDENTIALS` errors. Server requires NLA (`HYBRID_REQUIRED_BY_SERVER`) — expected.\n\n**Diagnosing any RDP auth failure: read the SERVER's journald, not the client error string.**\nClient errors all collapse different server causes into the same opaque message. Full server-log→cause\nmap, the `+auth-only`\u002F`\u002Fsec:tls` client probes, the cert\u002Flistener sanity checks, AND the\n**empty-server-log case** (tcpdump proves client-vs-network — e.g. a GUI client completing TCP then\nsending zero RDP bytes = client-side engine fault, fixed by recreating the connection, not touching\nthe server): `references\u002Frdp-auth-diagnosis-server-side.md`.\n\nFor **automated capture you do NOT need RDP** — Capability A is simpler. RDP is for a human.\nFull setup steps: `references\u002Fpeekaboo-and-rdp-setup-ubuntu-gnome-wayland.md` (Step 7).\n\n## Why the stock capture paths fail (so you don't waste time re-testing)\n\n1. `gnome-screenshot` WITHOUT the extension → exit 0 but no file (`AccessDenied` to the private API under the hood).\n2. Raw `gdbus`\u002F`busctl` call to `org.gnome.Shell.Screenshot` → `AccessDenied`. The extension allowlists the **`gnome-screenshot` binary's** D-Bus name, not arbitrary callers — so capture must go through the `gnome-screenshot` binary, not a hand-rolled gdbus call.\n3. `xdg-desktop-portal` `Screenshot` → no Response even with the PermissionStore granted (forced interactive dialog).\n4. `org.gnome.Mutter.ScreenCast.CreateSession` → `Session creation inhibited` (gated to grd itself).\n5. `ffmpeg kmsgrab` (KMS framebuffer) → \"No usable planes\" on NVIDIA proprietary KMS. `\u002Fdev\u002Ffb0` → all-black (GNOME doesn't scan out to fb0).\n6. `grim` → wlroots-only, fails on Mutter. Don't `apt install grim`.\n7. `enable unsafe_mode` via gdb injection into gnome-shell → **CRASHES the Shell, drops the desktop to the GDM greeter.** Do NOT use the gdb method. The extension is the safe, scoped, persistent unlock.\n\nNo usable **screen-coordinate** accessibility on Wayland — AT-SPI `getExtents(SCREEN)` is (0,0) on\nGTK4\u002FElectron. BUT this is **solved** by Capability B+ (`locate-element`): WINDOW-relative AT-SPI\nextents + Window Calls window origin → real click coords. Use `locate-element` for \"click element\nlabeled Save\"; fall back to vision-locate only where even WINDOW extents are zero.\n\n## Pitfalls\n\n- The screenshot unlock is **binary-scoped** — only the `gnome-screenshot` binary is allowlisted. A raw gdbus Screenshot call still fails. Always shell out to `gnome-screenshot`.\n- **Reloading the Shell on Wayland = restarting GDM** (`sudo systemctl restart gdm3`). There's no in-place `Alt+F2 r` over SSH. The **live GUI session is briefly torn down**, so don't do it mid-GUI-task. If autologin is enabled the desktop recovers in ~12s; `systemd --user` services survive if lingering is on. Verify recovery: `loginctl list-sessions` shows a `seat0` `Type=wayland Active=yes` session.\n- **Never use the gdb `unsafe_mode` injection** — it crashes gnome-shell. Use the extension.\n- **The `allow-gnome-screenshot` extension can silently flip to INACTIVE** (its allowlist hook only re-arms on a fresh Shell init). Symptom: `gnome-screenshot -f` exits 0 but writes no file. A disable\u002Fenable cycle is NOT enough — it needs a Shell reload (`sudo systemctl restart gdm3`) to re-arm. If you depend on capture in an automated loop, verify the file exists each time and reload-on-miss. (Window Calls can flip too — re-enable after any GDM restart.)\n- `gnome-screenshot -a` (interactive area) needs a human at the screen — useless headless; use `-f` (full) or `-w` (window).\n- Full-desktop shots are huge (multi-4K). Prefer `-w`, or downscale before analysis.\n- ydotool needs `ydotoold` alive + `YDOTOOL_SOCKET`\u002F`XDG_RUNTIME_DIR`. Re-enable if it dies. Raw `ydotool` is blind absolute pixels — prefer `locate-element` (Capability B+) for element targeting.\n- A **detached SSH session** isn't fully in the graphical session — export `XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID` (+ DBUS) explicitly. A native `systemd --user` service already has it.\n- Always LAN-scope RDP (3389) and SSH (22) in your firewall — never any-source.\n","# peekaboo-linux-gnome-wayland-computer-use\n\nThe Linux\u002FWayland counterpart to macOS computer-use — *\"see and poke the screen of a box for\ntroubleshooting or automation.\"* On macOS that's trivial (CGWindowList + CGEvent + AX tree). On\n**GNOME\u002FWayland it's locked down by design**, but it can be made to work end-to-end:\n\n- **See** → `gnome-screenshot` (re-enabled on GNOME 49+ via a small GNOME extension).\n- **Act** → `ydotool` (writes to `\u002Fdev\u002Fuinput`, no root).\n- **Target elements** → AT-SPI **WINDOW**-relative extents + the **Window Calls** extension's\n  window origin, composed into real absolute click coordinates (works where naive AT-SPI returns `(0,0)`).\n- **Hands-on (optional)** → GNOME Remote Desktop in Remote-Login mode for a human at an RDP client.\n\nBoth the \"see\" and \"act\" sides work **natively** (an agent process running ON the box) and\n**remotely** (from another machine over SSH). A native agent is a **single actor** — it can\nscreenshot, analyze, then ydotool-click in one loop. No second machine, no human, no RDP roundtrip.\n\nContext tags below — `[NATIVE]` = an agent process running ON the GNOME box (sharing the desktop\nsession of `user@1000`); `[REMOTE]` = on another machine, reaching the box over SSH.\n\n> Verified on Ubuntu 26.04 LTS + GNOME Shell 50 (Wayland), NVIDIA + AMD multi-GPU, multi-monitor.\n> Substitute your own host\u002Fuser where the examples use `USER@HOST` and `\u002Frun\u002Fuser\u002FUID`.\n\n## Capability matrix\n\n| Capability | Native (on box) | Remote (from another machine) |\n|---|---|---|\n| **Screenshot** | yes — `gnome-screenshot -f out.png` (extension unlocks it) | yes — same over SSH, then scp the file |\n| **Inject input** (mouse\u002Fkeys) | yes — ydotool via `\u002Fdev\u002Fuinput` (no root) | yes — over SSH, same ydotool |\n| **Click element by role\u002Fname** | yes — `locate-element` (AT-SPI WINDOW extents + Window Calls origin) | yes — same over SSH |\n| **Full see→reason→act loop** | yes — single actor, no RDP | yes — drive over SSH |\n| **Live interactive desktop** | n\u002Fa | yes — RDP client (optional, for a human) |\n\n## Target environment assumptions\n\n- A GNOME\u002FWayland box with a logged-in graphical session (the GUI session must be active on a seat).\n- For remote use: SSH access (key-based recommended), and the box's session env exported in\n  non-login shells — `XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID`,\n  `DBUS_SESSION_BUS_ADDRESS=unix:path=\u002Frun\u002Fuser\u002FUID\u002Fbus`, `WAYLAND_DISPLAY=wayland-0`, `DISPLAY=:0`.\n  (Replace `UID` with the session user's numeric id, e.g. `1000`.)\n- If agents run under `systemd --user` with the full graphical session env + lingering enabled,\n  they inherit all of the above and survive SSH disconnects.\n- `gnome-screenshot` grabs the **full virtual desktop** (all monitors stitched). On multi-4K\n  setups that's huge — use `-w` (focused window) or the per-display helper below.\n\n## CAPABILITY A — Screenshot (the key unlock)\n\nGNOME 49+ revoked `gnome-screenshot`'s access to the private Shell screenshot API, so on a stock\nbox every headless capture path is dead (see \"why\" below). The fix is the **`allow-gnome-screenshot`\nextension** (https:\u002F\u002Fgithub.com\u002Fsiddhpant\u002Fallow-gnome-screenshot, GNOME 49\u002F50): it re-adds\n`org.gnome.Screenshot` to the Shell's `_senderChecker._allowlistMap`, so the **`gnome-screenshot`\nbinary specifically** regains access. It is scoped to that one sender — it does NOT enable global\n`unsafe_mode`.\n\n`[NATIVE]` — just call it (session env already present):\n```bash\ngnome-screenshot -f \u002Ftmp\u002Fshot.png        # full virtual desktop (all monitors)\ngnome-screenshot -w -f \u002Ftmp\u002Fshot.png     # focused window only (smaller)\ngnome-screenshot -a -f \u002Ftmp\u002Fshot.png     # interactive area (needs a human at the screen — avoid headless)\n# then hand \u002Ftmp\u002Fshot.png to your image-analysis step\n```\n`[REMOTE]` — over SSH, then pull it back:\n```bash\nssh USER@HOST '\n  export XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID DBUS_SESSION_BUS_ADDRESS=unix:path=\u002Frun\u002Fuser\u002FUID\u002Fbus\n  gnome-screenshot -f \u002Ftmp\u002Fshot.png'\nscp USER@HOST:\u002Ftmp\u002Fshot.png \u002Ftmp\u002F   # then analyze it\n```\n\n**Sanity-check the capture** — a 4K full-desktop PNG is ~15-20 MB; a tiny\u002Fnear-empty file or an\nall-black image means the extension got disabled or the session dropped to the GDM greeter. Quick\nnon-black check:\n`python3 -c \"from PIL import Image; im=Image.open('\u002Ftmp\u002Fshot.png').convert('RGB'); d=list(im.getdata()); print(sum(1 for p in d[::997] if p!=(0,0,0)))\"` → should be well >0.\n\n### Installing the extension from scratch\n```bash\n# [NATIVE or over SSH with the session env exported]\nsudo apt-get install -y gnome-screenshot\nD=~\u002F.local\u002Fshare\u002Fgnome-shell\u002Fextensions\u002Fallow-gnome-screenshot@siddh.me\ngit clone --depth 1 https:\u002F\u002Fgithub.com\u002Fsiddhpant\u002Fallow-gnome-screenshot.git \u002Ftmp\u002Fags\nmkdir -p \"$D\" && cp -r \u002Ftmp\u002Fags\u002Fallow-gnome-screenshot@siddh.me\u002F* \"$D\u002F\"\nsudo systemctl restart gdm3      # reload the Shell so it discovers the extension (see WARNING below)\n# after the session returns:\nexport XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID DBUS_SESSION_BUS_ADDRESS=unix:path=\u002Frun\u002Fuser\u002FUID\u002Fbus\ngnome-extensions enable allow-gnome-screenshot@siddh.me\ngnome-extensions info allow-gnome-screenshot@siddh.me | grep State   # expect ACTIVE\n```\n\n### Per-display capture (capture ONE monitor, not the stitched canvas)\n\n`gnome-screenshot -f` grabs the **entire** virtual desktop. On multi-monitor setups that's a huge,\nmostly-empty canvas. To get a single clean display, use the bundled **`screenshot-display`** helper\n(`scripts\u002Fscreenshot-display.py`). It reads live monitor geometry from Mutter DisplayConfig and\ncrops the full capture to the chosen monitor.\n\n```bash\nscreenshot-display --list                               # enumerate monitors + geometry\nscreenshot-display --monitor primary --out \u002Ftmp\u002Fp.png   # primary (default)\nscreenshot-display --monitor DP-4   --out \u002Ftmp\u002Fm.png    # by connector name\nscreenshot-display --monitor 2      --out \u002Ftmp\u002Fm.png    # by index (0-based, sorted L->R, T->B)\nscreenshot-display --monitor active --out \u002Ftmp\u002Fw.png    # the FOCUSED WINDOW (best \"being worked on\" signal)\n```\nEach per-monitor crop is a clean single-display image — feed it straight to your vision step.\n`[REMOTE]`: run `screenshot-display` over SSH (export the session env), then scp the output back.\n\n### Window targeting — screenshot a specific program \u002F the display it's on\n\n`Shell.Eval`\u002F`Introspect` are locked, but the **Window Calls extension**\n(https:\u002F\u002Fgithub.com\u002Fickyicky\u002Fwindow-calls, GNOME 45-50) exposes window list + geometry over D-Bus\nat `\u002Forg\u002Fgnome\u002FShell\u002FExtensions\u002FWindows`. The `screenshot-display` helper uses it to find and crop\nto a window or its monitor:\n\n```bash\nscreenshot-display --list-windows                       # wm_class, id, focus, monitor, title\nscreenshot-display --window focused      --out \u002Ftmp\u002Fw.png   # crop to focused window's exact rect\nscreenshot-display --window \"wm_class=firefox\" --out \u002Ftmp\u002Fw.png  # match by wm_class substring\nscreenshot-display --window \"title=Settings\"   --out \u002Ftmp\u002Fw.png  # match by title substring\nscreenshot-display --window-display focused        --out \u002Ftmp\u002Fd.png  # the whole MONITOR that window sits on\nscreenshot-display --window-display \"wm_class=Code\" --out \u002Ftmp\u002Fd.png  # monitor a matched window is on\n```\nFlow: `List` → pick window (focused \u002F wm_class \u002F title) → `Details` gives `x,y,width,height,monitor`\n→ helper crops the full capture to the window rect, or to whichever monitor's rect contains the\nwindow's center. (`Details` works; `GetFrameBounds` is broken in this ext version — the helper\ndoesn't use it.)\n\n**Full setup\u002Fenablement guide** to replicate on any GNOME\u002FWayland box:\n`references\u002Fpeekaboo-and-rdp-setup-ubuntu-gnome-wayland.md`.\n\n## CAPABILITY B — Input injection (ydotool)\n\nydotool writes to `\u002Fdev\u002Fuinput` below the display server (the input analog of the screenshot\nunlock). Install it, run `ydotoold` as a persistent user service (socket\n`\u002Frun\u002Fuser\u002FUID\u002F.ydotool_socket`), and ACL `\u002Fdev\u002Fuinput` to your user (no root needed at runtime).\n\n`[NATIVE]`:\n```bash\nexport YDOTOOL_SOCKET=\u002Frun\u002Fuser\u002FUID\u002F.ydotool_socket\nydotool mousemove -a 960 540     # absolute pixel coords (no element targeting)\nydotool click 0xC0               # left click down+up\nydotool type \"hello\"\nydotool key 28:1 28:0            # Enter (keycode:press \u002F keycode:release)\n```\n`[REMOTE]`:\n```bash\nssh USER@HOST '\n  export XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID YDOTOOL_SOCKET=\u002Frun\u002Fuser\u002FUID\u002F.ydotool_socket\n  ydotool mousemove -a 700 400; ydotool click 0xC0; ydotool type \"hi\"'\n```\nIf `ydotoold.service` is down — `systemctl --user enable --now ydotoold.service`. For element\ntargeting (not blind coords) use the `locate-element` helper below.\n\n## CAPABILITY B+ — Click an element by role\u002Fname (SOM via hybrid coords)\n\n**You do NOT need blind pixel guessing or a vision round-trip for most clicks.** The bundled\n**`locate-element`** helper (`scripts\u002Flocate-element.py`) resolves a UI element to a real absolute\ncoordinate and clicks\u002Ftypes it.\n\n**Why it works when AT-SPI looks broken:** on Wayland, `getExtents(SCREEN)` returns `(0,0)` for\nGTK4\u002FElectron (no global coordinates by design), so naive AT-SPI clicking fails. But\n`getExtents(WINDOW)` (widget-relative) **is** populated, and **Window Calls** gives the window's\nscreen origin from Mutter's privileged view. The helper composes them:\n`abs = window_origin + window_relative_extents + size\u002F2`. Proven on GTK4 apps.\n\n```bash\nlocate-element --app gnome-text-editor --list                       # list elements + computed ABS coords\nlocate-element --app gnome-text-editor --role text --click          # click first text widget\nlocate-element --app org.gnome.Nautilus --name \"New Folder\" --click  # click by visible label\nlocate-element --app gnome-text-editor --role text --click --type \"hello\"   # click then type\nlocate-element --window-id 1367000216 --role \"push button\" --name Open --click  # by explicit window id\nlocate-element --app firefox --name Reload --print                  # just print \"X Y\", act yourself\n```\n- `--app` matches the AT-SPI app name AND wm_class loosely (normalizes `org.gnome.TextEditor` ↔ `gnome-text-editor`).\n- `--role`\u002F`--name` are case-insensitive substring filters; `--index N` picks the Nth match.\n- Needs `toolkit-accessibility=true` (`gsettings set org.gnome.desktop.interface toolkit-accessibility true`),\n  Window Calls active, and ydotoold running.\n\n**Vision fallback (only when an element zeros even WINDOW extents** — custom-drawn canvases, some\nElectron internals): `locate-element` exits non-zero and prints the fallback recipe —\n`screenshot-display --window 'wm_class=X' --out \u002Ftmp\u002Fw.png` → analyze for pixel coords →\n`ydotool mousemove -a -x X -y Y`.\n\nDeep dive on the technique, the gotchas, and why Newton isn't a substitute yet:\n`references\u002Fwayland-atspi-hybrid-som.md`.\n\n## The native loop (single actor — the whole point)\n\n```\n[NATIVE agent on the box]\n  1. gnome-screenshot -f \u002Ftmp\u002Fs.png\n  2. analyze \u002Ftmp\u002Fs.png  → \"the Save button is at ~(1820, 640)\"\n  3. ydotool mousemove -a 1820 640 && ydotool click 0xC0\n  4. gnome-screenshot again → verify it landed\n```\nNo second machine, no human, no RDP. For element targeting prefer `locate-element` over blind coords.\n\n## Window actions (focus \u002F move \u002F resize \u002F close)\n\nWindow Calls also does focus\u002Fmove\u002Fresize\u002Fclose verbs (window id from `--list-windows`):\n```bash\nWIN=\u002Forg\u002Fgnome\u002FShell\u002FExtensions\u002FWindows\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Activate \u003Cid>   # focus\u002Fraise\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Move \u003Cid> \u003Cx> \u003Cy>\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Resize \u003Cid> \u003Cw> \u003Ch>\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Maximize \u003Cid>\ngdbus call --session --dest org.gnome.Shell --object-path $WIN --method $WIN.Close \u003Cid>\n```\nFlow: `Activate \u003Cid>` → `screenshot-display --window focused` → `ydotool` interact.\n\n## CAPABILITY C — Live interactive desktop via RDP (optional, for a human)\n\nFor a person to drive the desktop hands-on from another machine, use **GNOME Remote Desktop** in\n**Remote Login** mode (system\u002Fheadless, fresh negotiated display — NOT Desktop Sharing, which\nmirrors the physical multi-monitor desktop and re-hits the huge-canvas problem).\n\n- Mode: `grdctl --system` (system service `gnome-remote-desktop.service`, enabled at boot).\n- Listener: `*:3389` — **always LAN-scope it in your firewall, never any-source.**\n- TLS cert at `\u002Fvar\u002Flib\u002Fgnome-remote-desktop\u002Frdp-tls.{crt,key}`.\n- Credentials: a device user + password over NLA. Setting the device cred = the Linux user's\n  password makes both stages (NLA device prompt, then GDM login) take the same credential.\n\nConnect from a client (Microsoft Remote Desktop \u002F Windows App, or any RDP client) → `HOST` → device\nuser\u002Fpass → GDM → sign in as the Linux user.\n\n**⚠️ #1 RDP gotcha — credential changes need a daemon restart.** `grdctl --system rdp set-credentials`\nwrites the keyfile but the **running daemon does NOT hot-reload it into its NTLM SAM**. Until you\n`sudo systemctl restart gnome-remote-desktop.service`, NLA fails for every client:\n`ntlm_fetch_ntlm_v2_hash: Could not find user in SAM database` → `SEC_E_NO_CREDENTIALS` →\n`transport_accept_nla: client authentication failure`. Meanwhile `grdctl status` shows the correct\ncreds (it reads the file). Clients report this as **`FREERDP_ERROR_CONNECT_TRANSPORT_FAILED`** (RoyalTS)\nor `ERRCONNECT_AUTHENTICATION_FAILED` (FreeRDP) — looks like transport\u002Fcert, is actually stale creds.\nFix: set creds (password via STDIN) → **restart the service** → verify journald shows no `SAM`\u002F\n`NO_CREDENTIALS` errors. Server requires NLA (`HYBRID_REQUIRED_BY_SERVER`) — expected.\n\n**Diagnosing any RDP auth failure: read the SERVER's journald, not the client error string.**\nClient errors all collapse different server causes into the same opaque message. Full server-log→cause\nmap, the `+auth-only`\u002F`\u002Fsec:tls` client probes, the cert\u002Flistener sanity checks, AND the\n**empty-server-log case** (tcpdump proves client-vs-network — e.g. a GUI client completing TCP then\nsending zero RDP bytes = client-side engine fault, fixed by recreating the connection, not touching\nthe server): `references\u002Frdp-auth-diagnosis-server-side.md`.\n\nFor **automated capture you do NOT need RDP** — Capability A is simpler. RDP is for a human.\nFull setup steps: `references\u002Fpeekaboo-and-rdp-setup-ubuntu-gnome-wayland.md` (Step 7).\n\n## Why the stock capture paths fail (so you don't waste time re-testing)\n\n1. `gnome-screenshot` WITHOUT the extension → exit 0 but no file (`AccessDenied` to the private API under the hood).\n2. Raw `gdbus`\u002F`busctl` call to `org.gnome.Shell.Screenshot` → `AccessDenied`. The extension allowlists the **`gnome-screenshot` binary's** D-Bus name, not arbitrary callers — so capture must go through the `gnome-screenshot` binary, not a hand-rolled gdbus call.\n3. `xdg-desktop-portal` `Screenshot` → no Response even with the PermissionStore granted (forced interactive dialog).\n4. `org.gnome.Mutter.ScreenCast.CreateSession` → `Session creation inhibited` (gated to grd itself).\n5. `ffmpeg kmsgrab` (KMS framebuffer) → \"No usable planes\" on NVIDIA proprietary KMS. `\u002Fdev\u002Ffb0` → all-black (GNOME doesn't scan out to fb0).\n6. `grim` → wlroots-only, fails on Mutter. Don't `apt install grim`.\n7. `enable unsafe_mode` via gdb injection into gnome-shell → **CRASHES the Shell, drops the desktop to the GDM greeter.** Do NOT use the gdb method. The extension is the safe, scoped, persistent unlock.\n\nNo usable **screen-coordinate** accessibility on Wayland — AT-SPI `getExtents(SCREEN)` is (0,0) on\nGTK4\u002FElectron. BUT this is **solved** by Capability B+ (`locate-element`): WINDOW-relative AT-SPI\nextents + Window Calls window origin → real click coords. Use `locate-element` for \"click element\nlabeled Save\"; fall back to vision-locate only where even WINDOW extents are zero.\n\n## Pitfalls\n\n- The screenshot unlock is **binary-scoped** — only the `gnome-screenshot` binary is allowlisted. A raw gdbus Screenshot call still fails. Always shell out to `gnome-screenshot`.\n- **Reloading the Shell on Wayland = restarting GDM** (`sudo systemctl restart gdm3`). There's no in-place `Alt+F2 r` over SSH. The **live GUI session is briefly torn down**, so don't do it mid-GUI-task. If autologin is enabled the desktop recovers in ~12s; `systemd --user` services survive if lingering is on. Verify recovery: `loginctl list-sessions` shows a `seat0` `Type=wayland Active=yes` session.\n- **Never use the gdb `unsafe_mode` injection** — it crashes gnome-shell. Use the extension.\n- **The `allow-gnome-screenshot` extension can silently flip to INACTIVE** (its allowlist hook only re-arms on a fresh Shell init). Symptom: `gnome-screenshot -f` exits 0 but writes no file. A disable\u002Fenable cycle is NOT enough — it needs a Shell reload (`sudo systemctl restart gdm3`) to re-arm. If you depend on capture in an automated loop, verify the file exists each time and reload-on-miss. (Window Calls can flip too — re-enable after any GDM restart.)\n- `gnome-screenshot -a` (interactive area) needs a human at the screen — useless headless; use `-f` (full) or `-w` (window).\n- Full-desktop shots are huge (multi-4K). Prefer `-w`, or downscale before analysis.\n- ydotool needs `ydotoold` alive + `YDOTOOL_SOCKET`\u002F`XDG_RUNTIME_DIR`. Re-enable if it dies. Raw `ydotool` is blind absolute pixels — prefer `locate-element` (Capability B+) for element targeting.\n- A **detached SSH session** isn't fully in the graphical session — export `XDG_RUNTIME_DIR=\u002Frun\u002Fuser\u002FUID` (+ DBUS) explicitly. A native `systemd --user` service already has it.\n- Always LAN-scope RDP (3389) and SSH (22) in your firewall — never any-source.\n",{"name":5,"metadata":24,"description":7},{"clawdbot":25},{"os":26,"emoji":29},[27,28],"darwin","linux","🖥️","master","4aa004da461d820f4ea5499e58eb8d196cf99016","community\u002Fclawhub\u002Fp\u002Fpeekaboo-linux-gnome-wayland-computer-use",0,"https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge","https:\u002F\u002Fskill-explorer-rho.vercel.app","https:\u002F\u002Fgithub.com\u002FLord1Egypt","Unbreakable. Unstoppable. Unmatched⚡.\r\n Speaker of machine tongues 🤖  | World-builder of decentralized dreams 🌐 | Defender of the digital realm🛡️.","2026-06-10 08:21:01",{"source":8,"skillName":5,"commands":40,"prompt":50,"zipAvailable":51},[41,44,47],{"tool":42,"command":43},"npx","npx skills add Lord1Egypt\u002Fawesome-skill-forge --skill peekaboo-linux-gnome-wayland-computer-use",{"tool":45,"command":46},"bunx","bunx skills add Lord1Egypt\u002Fawesome-skill-forge --skill peekaboo-linux-gnome-wayland-computer-use",{"tool":48,"command":49},"pnpm","pnpm dlx skills add Lord1Egypt\u002Fawesome-skill-forge --skill peekaboo-linux-gnome-wayland-computer-use","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fcommunity\u002Fclawhub\u002Fp\u002Fpeekaboo-linux-gnome-wayland-computer-use\u002FSKILL.md",true,[],[54,61,68,75,82,89],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M785WKJB99G3STR1NF9NRTJW","youtube-content","YouTube transcripts to summaries, threads, blogs.","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fskills\u002Fyoutube-content\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M785WEFHB0G2S9YA6EPR3H0E","xurl","X\u002FTwitter via xurl CLI: post, search, DM, media, v2 API.","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fskills\u002Fxurl\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M785W98EDN4EFJXJPMQEHCRC","touchdesigner-mcp","Control a running TouchDesigner instance via twozero MCP — create operators, set parameters, wire connections, execute Python, build real-time visuals. 36 native tools.","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fskills\u002Ftouchdesigner-mcp\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M785W5S9FP3QBH2V191E025P","test-driven-development","TDD: enforce RED-GREEN-REFACTOR, tests before code.","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fskills\u002Ftest-driven-development\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M785W2AS08NSSH9YATCV6FJB","teams-meeting-pipeline","Operate the Teams meeting summary pipeline via Hermes CLI — summarize meetings, inspect pipeline status, replay jobs, manage Microsoft Graph subscriptions.","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fskills\u002Fteams-meeting-pipeline\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M785VXKW4T34BTP1SCHW1R7F","systematic-debugging","4-phase root cause debugging: understand bugs before fixing.","https:\u002F\u002Fgithub.com\u002FLord1Egypt\u002Fawesome-skill-forge\u002Fblob\u002Fmaster\u002Fskills\u002Fsystematic-debugging\u002FSKILL.md",[],[]]