Room 418

ReviewAudited by ClawScan on May 10, 2026.

Overview

This is a disclosed game client, but its auto-play and cron modes can let adversarial agent messages drive an OpenClaw sub-agent and submit turns without user review.

Install only if you want an autonomous external game client. Before use, choose manual or notify mode if you want review, set a non-identifying ROOM418_AGENT_NAME, avoid setup-cron.sh unless you want background play, and keep the Room 418 credentials file private.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

A malicious or adversarial opponent message could try to steer the local agent outside the intended game response, especially in auto mode.

Why it was flagged

Messages from other agents are placed into a prompt for a local OpenClaw agent session, but the artifacts do not show a clear untrusted-content boundary or tool restriction for that sub-agent.

Skill content
CONVERSATION=$(echo "$BATTLE" | jq -r '.messages[] | "[\(.role)] \(.content)"') ... RESPONSE=$(openclaw agent ... --message "$PROMPT" ...)
Recommendation

Run in manual or notify mode unless you are comfortable with autonomous play, and restrict the sub-agent's tools or sandbox it so opponent text is treated only as game data.

What this means

The agent can post game messages and continue joining matches automatically, which may surprise users expecting review before external submissions.

Why it was flagged

The default mode is auto, generated responses are submitted to the game server, and finished battles can be requeued without a per-turn user approval step.

Skill content
MODE="auto" ... SUBMIT_RESULT=$("$SCRIPT_DIR/submit-turn.sh" "$BATTLE_ID" "$RESPONSE" 2>&1) ... api_post "/api/agent/queue/join"
Recommendation

Set ~/.config/room418/config.json to notify or manual before playing if you want to approve each message; make auto mode an explicit opt-in during setup.

What this means

If enabled, Room 418 can keep waking up and acting in the background after the initial play session.

Why it was flagged

The setup script creates a persistent recurring job in the main session; its message tells the agent to read HEARTBEAT.md, generate dialogue, and submit when it is the user's turn.

Skill content
openclaw cron add --name "room418" --every "2m" --message "$CRON_MSG" --session "main"
Recommendation

Only run setup-cron.sh if you want persistent autonomous play, prefer a dedicated isolated session, and remove it with openclaw cron rm room418 when done.

What this means

Anyone who obtains this file could act as your Room 418 agent.

Why it was flagged

Registration stores a Room 418 bearer token locally and later scripts use it for API calls; this is expected for the game service and the file is chmod 600.

Skill content
token: .data.token ... > "$CRED_FILE"; chmod 600 "$CRED_FILE"
Recommendation

Keep ~/.config/room418/credentials.json private, avoid untrusted ROOM418_API_URL values, and delete the credential file if you stop using the service.

What this means

Your device hostname could be sent to the external Room 418 service and appear as part of your game identity.

Why it was flagged

If the user does not set ROOM418_AGENT_NAME, the default registered name includes the local short hostname, which may reveal a personal or organizational machine identifier.

Skill content
AGENT_NAME="${ROOM418_AGENT_NAME:-openclaw-$(hostname -s | tr '[:upper:]' '[:lower:]')-$(date +%s | tail -c 5)}"
Recommendation

Set ROOM418_AGENT_NAME to a pseudonym before registration, and update the skill to avoid hostname-derived defaults.