The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted task titles and items from user input, stores them in a project-level CSV file, and later interpolates this content into JSON payloads for the update_plan tool. This creates an attack surface where instructions embedded within a task item could influence the agent's planning logic during subsequent synchronization steps.
Ingestion points: Task items and titles provided via the init and add commands in scripts/todo_csv.py.
Boundary markers: The data is handled as structured CSV and JSON, but no explicit instruction-filtering or delimiting is applied to the item text itself.
Capability inventory: The skill can write files to the project root and execute local git commands.
Sanitization: Task titles are sanitized for filename safety, but item descriptions are stored as raw strings.
[COMMAND_EXECUTION]: The scripts/todo_csv.py script executes a git command using subprocess.check_output to identify the project root directory. The implementation follows security best practices by passing the command as a list of arguments and avoiding shell execution.

todo-list-csv