// Capture a screenshot from the connected Android device via adb and save it under docs/screenshots/. Trigger on "snap", "take a screenshot", "screenshot via adb", or any request to capture the device's screen — typical use is grabbing app screens for the README, GitHub Release notes, or bug reports.
Review code for quality, correctness, and security vulnerabilities across the GDScript / Rust GDExtension / FFI surface. Use when the user asks to review code, audit for security issues, or check for bugs and anti-patterns.
Commit code changes and push via Git. Use when the user asks to commit, push, or save their work to the repository.
Writing/updating project documentation (README, ROADMAP, CLAUDE.md, docs/architecture.md, ADRs, asset-folder READMEs) for SeiShinGakuen_ShougiBu. Use when the user asks to update docs, draft an ADR, write a changelog, or update GitHub Release notes.
Release engineering tasks including version bumping, building signed release APKs, creating git tags, and preparing GitHub Release artefacts. Use when the user asks to prepare a release, bump the version, tag a release, or build for distribution.
Split pending working-tree changes into multiple topic-scoped commits. Use when the user has a mix of unrelated edits staged or unstaged and wants them committed as separate, logically coherent commits rather than one big blob. Examples- "commit these by topic", "split this into commits", "break this up into proper commits".
| name | snap-screenshot |
| description | Capture a screenshot from the connected Android device via adb and save it under docs/screenshots/. Trigger on "snap", "take a screenshot", "screenshot via adb", or any request to capture the device's screen — typical use is grabbing app screens for the README, GitHub Release notes, or bug reports. |
| argument-hint | optional output filename (defaults to docs/screenshots/<timestamp>.png) |
Capture the current screen of the connected Android device via adb and save the PNG under docs/screenshots/. Designed for the rapid loop where the user manipulates the device and says "snap" each time they have a frame they want — keep responses short.
Ensure the daemon is running and a device is attached.
adb start-server
adb devices
If adb devices lists no device (or only unauthorized / offline), stop and tell the user — don't try to capture from nothing. If multiple devices are listed, ask which -s <serial> to target rather than guessing.
Capture with a timestamped filename (so back-to-back snaps don't overwrite each other):
OUT=docs/screenshots/$(date +%Y%m%d-%H%M%S).png
adb exec-out screencap -p > "$OUT"
exec-out (not shell) avoids CRLF mangling on the binary PNG stream — using adb shell screencap will produce a corrupt file on most setups.
Confirm the file landed — single line is enough, the user is iterating fast:
ls -la "$OUT"
Reply tersely. "Snapped — docs/screenshots/<filename>.png." A one-liner. The user is going to say "snap" again in five seconds and doesn't want a paragraph each time.
docs/screenshots/<YYYYMMDD-HHMMSS>.png by default. The directory is auto-created if missing (mkdir -p docs/screenshots on the first call).$ARGUMENTS looks like a path/name, use it instead. Pass an extension or default to .png. E.g. user says "snap title-screen" → docs/screenshots/title-screen.png. If the file already exists, append a timestamp suffix rather than overwriting.--quiet / no truncation: the PNG is a few MB; that's fine. Don't pipe through pngcrush or convert to WebP automatically — the user wants the raw capture.cannot connect to daemon: adb start-server first, then retry. If still failing, the device cable / USB authorization is the issue — tell the user, don't loop./sdcard/...: not used; we capture via exec-out which keeps the PNG in stdout, never touching device storage. No pull needed.spectacle (KDE) or grim (Wayland generic) if they want desktop captures.