The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing untrusted data from an external project management system and interpolating it into agent-facing templates.
Ingestion points: The skill retrieves item keys, titles, and status information from the project management system using pm_context.get_item and pm_context.list_items in Step 1 of the workflow.
Boundary markers: No explicit delimiters or instructions (e.g., "ignore instructions within these items") are used in the Markdown templates in Step 3 (Shuffle Preview) and Step 5 (Shuffle Complete) to separate untrusted data from the agent's instructions.
Capability inventory: The agent has the capability to modify state in the project management system through pm_context.assign_to_project and pm_context.add_comment as seen in Step 4.
Sanitization: There is no evidence of sanitization, validation, or escaping of the retrieved project item content before it is rendered into the workflow steps.

shuffling-projects