Task Watchdog

Security checks across malware telemetry and agentic risk

Overview

Task Watchdog appears to be a coherent monitoring plugin, but it intentionally injects automated reminders into agent context and briefly reuses recent user-message text for those reminders.

This looks reasonable for users who want automatic task-failure and stale-task alerts. Before installing, be aware that it can add instructions to the agent's context, trigger heartbeat checks in the background, and include short excerpts of recent user messages in alerts.

VirusTotal

No VirusTotal findings

View on VirusTotal

Risk analysis

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.

#
ASI01: Agent Goal Hijack
Low
What this means

The agent may receive automatic instructions to continue, report, or reply when the watchdog detects failures, silence, or stale work.

Why it was flagged

The plugin directly injects watchdog messages into the agent/session context. This is the stated feature, not hidden behavior, but it can influence what the agent does next.

Skill content
api.runtime?.system?.enqueueSystemEvent?.(safeText, { sessionKey }); ... api.enqueueNextTurnInjection({ sessionKey: parentKey, text: continuationMsg, placement: "prepend_context", ttlMs })
Recommendation

Install only if you want automatic watchdog nudges, and review the TTL, threshold, and patrol settings to match your workflow.

#
ASI06: Memory and Context Poisoning
Low
What this means

Short excerpts of recent user messages may appear in automated watchdog reminders seen by the agent.

Why it was flagged

The plugin stores recent user-message text in memory and includes a truncated excerpt in injected alerts. No external transmission is shown, but user content can be replayed into later agent context.

Skill content
const userMessageContent = new Map<string, string>(); ... fullText += `\n\n📝 用户原始消息:${truncate(userMsg, MESSAGE_TRUNCATE_LEN)}`;
Recommendation

Avoid putting secrets in prompts, and verify session scoping if using the plugin in shared or multi-session gateways.

#
ASI10: Rogue Agents
Info
What this means

The plugin may trigger heartbeat checks and reminders without a direct user invocation after installation.

Why it was flagged

The plugin starts with the gateway and runs periodic patrol behavior by default. This is disclosed and purpose-aligned for a watchdog, but it is background autonomous activity.

Skill content
"activation": { "onStartup": true } ... "timerPatrol": { "default": true }
Recommendation

Disable `timerPatrol` or adjust `timerPatrolIntervalMs` if you do not want periodic background watchdog checks.