The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes commands to modify system schedulers (crontab and launchd).
Evidence: Scripts add-to-crontab.sh and add-to-launchd.sh directly interface with OS scheduling utilities to register recurring tasks.
Mitigation: The skill explicitly requires interactive user confirmation before applying any changes to the system scheduler and uses helper scripts to ensure consistent, validated behavior.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by scheduling tasks that may process external data.
Ingestion points: Task prompts are generated based on user requests which could include data from files, web pages, or other MCP tools.
Boundary markers: No explicit boundary markers or 'ignore' instructions are enforced in the generated task prompts.
Capability inventory: The skill can execute arbitrary prompts via opencode or kiro CLI tools in a headless background environment.
Sanitization: There is no evidence of sanitization or escaping of the prompt content before it is written to the scheduler configuration files.

task-trigger