The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/search_content.sh and scripts/search_by_tag.sh pass user-controlled variables directly to grep commands. This enables argument injection (e.g., using -f to read patterns from files), allowing an attacker to manipulate the search tool's behavior.\n- [DATA_EXFILTRATION]: The scripts accept a directory path as the first argument (KANBAN_DIR) without performing any validation or sandboxing. This allows the skill to be used to read and process files in sensitive directories outside the intended workspace if the agent is directed to do so.\n- [INDIRECT_PROMPT_INJECTION]:\n
Ingestion points: Markdown files located in the kanban/ directory are read and processed by the shell scripts.\n
Boundary markers: Absent. There are no markers to distinguish between data and instructions when the script output is returned to the agent.\n
Capability inventory: The skill uses bash to execute scripts that call grep, awk, and sed for file access and text processing.\n
Sanitization: Absent. Content from the Markdown cards is displayed directly to the agent without filtering or escaping.

kanban-ai