The Agent Skills Directory

COMMAND_EXECUTION (SAFE): The skill executes a local shell command cat ~/.claude/daily-tasks/$(date +%Y-%m-%d).json to retrieve task data. The path is constrained to a specific sub-directory and uses a system-generated date for the filename, minimizing injection risks.
DATA_EXPOSURE & EXFILTRATION (SAFE): While the skill accesses files within a hidden directory in the user's home folder (~/.claude/), the access is localized to task management data. There are no network operations (curl, wget, or fetch) that would indicate data exfiltration.
INDIRECT PROMPT INJECTION (LOW): The skill ingests untrusted data from a JSON file and user-provided arguments (the 'reason' for standby).
Ingestion points: ~/.claude/daily-tasks/*.json.
Boundary markers: None explicitly mentioned in the instructions for the JSON parsing.
Capability inventory: Uses cat for reading and an internal 'Write tool' for updating files.
Sanitization: No explicit sanitization of the 'reason' string is mentioned before it is written back to the JSON or displayed to the user. While this is a potential surface for indirect injection if other tools read this data, the current skill's capabilities are limited and do not pose a high risk.

daily-standby