marketing-screenshot
Skill22 starsChanged 46 days agoMITView on GitHub
Capture the real Gravity desktop app for the marketing site using a disposable daemon and staged example data. Real Claude Code sessions are billed and require explicit consent.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- bus-live-testSkill
- releaseSkill
- terminal-perf-testSkill
- visual-regressionSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 7 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 8
Use the native macOS app with fictional project data.
- Very specific: line 13
Keep the installed app and daemon untouched.
- Distinctive: line 21
Run from the repository root, in a dedicated terminal.
- Distinctive: line 24
Keep shell tracing off when loading credentials.
- Very specific: line 49
Wait for $GRAVITY_HOME/gravityd.port and a successful curl --fail http://127.0.0.1:49888/health.
- Distinctive: line 54
Start the app in the same terminal.
- Distinctive: line 67
If it differs, stop and reset only that disposable profile's connection settings.
- Distinctive: line 68
Never install, restart, or update the managed production daemon from this app.
- Distinctive: line 69
Do not expose the dev frontend to the network or deploy a build containing its local test token.
- Distinctive: line 98
Use available native accessibility/window tools.
- Very specific: line 119
Stop the foreground pnpm tauri dev with Ctrl-C.
- Very specific: line 121
If the shell/session was lost, verify process ownership again before signalling.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Marketing screenshot
Use the native macOS app with fictional project data. Confirm consent and a
small turn budget before starting runtime = "pty": creating each bot starts
a real, billed Claude Code session. Use double for a free layout rehearsal;
it cannot demonstrate genuine agent behavior.
Keep the installed app and daemon untouched. Never assume their port or data
location; use a fresh daemon home and an unused test port. Requirements are in
CONTRIBUTING.md; the driver additionally requires Node 22+. A real run needs
an authenticated claude executable on PATH. Do not print credentials, raw
process command lines, or real session transcripts into shared reports.
Setup
Run from the repository root, in a dedicated terminal. These examples use 49888 for the daemon and 1420 for the frontend. Check both are free first; choose other ports consistently if occupied, without stopping their owners. Keep shell tracing off when loading credentials.
umask 077
SHOT_DIR=$(mktemp -d "${TMPDIR:-/tmp}/gravity-shot.XXXXXX")
SHOT_DIR=$(cd "$SHOT_DIR" && pwd -P) # canonical paths also avoid transcript aliases
export GRAVITY_HOME="$SHOT_DIR/daemon"
export GRAVITY_WS=ws://127.0.0.1:49888/ws
mkdir -p "$GRAVITY_HOME" "$SHOT_DIR/shots"
CLAUDE_BIN=$(command -v claude)
cargo build -p gravityd
./scripts/prepare-sidecar.sh
cat > "$GRAVITY_HOME/gravityd.toml" <<EOF
home = "$GRAVITY_HOME"
bind = ["127.0.0.1"]
port = 49888
negotiate_port = false
runtime = "pty"
claude_bin = "$CLAUDE_BIN"
supervision_interval_ms = 1000
EOF
./target/debug/gravityd --config "$GRAVITY_HOME/gravityd.toml" > "$SHOT_DIR/gravityd.log" 2>&1 &
SHOT_DAEMON_PID=$!Wait for $GRAVITY_HOME/gravityd.port and a successful
curl --fail http://127.0.0.1:49888/health. Confirm the recorded daemon PID
is still running and the published port is 49888. Do not continue on a startup
failure. Leave $HOME unchanged so Claude Code can use its existing login.
Start the app in the same terminal; the dev token stays in the local process environment, not a command argument or a pasted browser-tool call:
export VITE_GRAVITY_DEV_PORT=49888
export VITE_GRAVITY_DEV_TOKEN="$(cat "$GRAVITY_HOME/secrets/client.token")"
(cd apps/desktop && env -u CONDUCTOR_PORT pnpm tauri dev \
--config '{"identifier":"build.getgravity.screenshot","build":{"devUrl":"http://localhost:1420"}}')The separate app identifier separates screenshot app storage from the official
app. A previous screenshot profile can still contain connection overrides:
the footer must read 127.0.0.1:49888 · connected before any interaction.
If it differs, stop and reset only that disposable profile's connection settings.
Never install, restart, or update the managed production daemon from this app.
Do not expose the dev frontend to the network or deploy a build containing its
local test token.
Staging the scene
In another terminal, set GRAVITY_HOME to the exact disposable daemon home
from setup and GRAVITY_WS to its loopback URL. The bundled driver validates
that the endpoint matches the home's published port before reading credentials.
DRIVER=.claude/skills/marketing-screenshot/scripts/drive.mjs
PROJECT_ID=$(node "$DRIVER" project Example)
node "$DRIVER" bot "$PROJECT_ID" Argus "Coordinates the example project." \
"Use fictional example data only. Create no additional bots until asked." icon:quartzThe driver prints the bot ID. Use say <bot-id> <prompt> to ask for a small,
budgeted task, or additional bots within the approved budget. list shows IDs
and states; peek <bot-id> prints terminal content, so review it locally for
sensitive information before sharing it. update changes a bot's charter.
delete archives a bot: identify the exact bot and obtain deletion approval
before using it. Every command must target this run's disposable home.
Let the work settle naturally. Do not backdate database messages or add symlinks inside the user's Claude transcript directory to make the scene look busy. Pin the lead bot, select it, and collapse the info panel if that suits the shot.
Capturing
Use available native accessibility/window tools. Identify the dev app's exact
PID and executable path from the session you launched; do not target a process
by name alone. The installed app may have the same executable name. Inspect
only the chosen window, raise it, and read its current bounds. For example,
with the verified PID in SHOT_APP_PID:
osascript -e "tell application \"System Events\" to tell (first process whose unix id is $SHOT_APP_PID) to perform action \"AXRaise\" of window 1"
osascript -e "tell application \"System Events\" to tell (first process whose unix id is $SHOT_APP_PID) to get {position, size} of window 1"
# Set SHOT_BOUNDS to those four numbers as x,y,width,height, then:
screencapture -o -x -R "$SHOT_BOUNDS" "$SHOT_DIR/shots/gravity-hero.png"
sips -g pixelWidth -g pixelHeight "$SHOT_DIR/shots/gravity-hero.png"Resolution depends on the display scale and actual window size. Inspect the saved PNG for overlapping windows, private paths, account information and real project data before copying it into public marketing assets. Do not claim synthetic output is a real agent session.
Cleanup
Stop the foreground pnpm tauri dev with Ctrl-C. Stop only the daemon PID
recorded in this live shell (kill "$SHOT_DAEMON_PID"; wait "$SHOT_DAEMON_PID").
If the shell/session was lost, verify process ownership again before signalling;
never use broad pkill patterns. Unset the temporary Vite token environment.
Retain the screenshot and run directory for review. Ask before deleting the
exact directory. Real Claude sessions may also create transcript directories
under ~/.claude/projects and trust entries in ~/.claude.json. Report this
residue; do not delete globbed paths or rewrite the user's configuration.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

