The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly states that a Telegram bot token is embedded directly within the scripts/lib/alert-bus.sh script, which is a significant security risk.
[CREDENTIALS_UNSAFE]: The skill references and accesses sensitive OAuth2 credential files (credentials.json and token.json) stored in skills/skippy-google/.
[DATA_EXFILTRATION]: The skill provides a command format [Radar] Deliver doc: <path> which allows the agent to read arbitrary local files and send them to external channels like Telegram or Gmail. This capability can be abused to exfiltrate sensitive system files.
[COMMAND_EXECUTION]: The skill utilizes a systemd timer via scripts/radar-dispatch.sh to maintain persistence and perform background polling every 5 minutes.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its polling of the TASKS.md file. Any external input or other skill that can write to this file can trigger the Radar skill to perform unauthorized actions or exfiltrate data.
Ingestion points: TASKS.md (specified in SKILL.md).
Boundary markers: Uses the [Radar] prefix, which is a weak delimiter and easily spoofable.
Capability inventory: Includes Bash, Read, Write, and Edit tools, along with scripts to send data to Telegram and Gmail.
Sanitization: There is no evidence of sanitization or path validation for the <path> argument in document delivery commands.

radar