The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and process raw document content (e.g., the first 2000 characters of a file) to determine if it matches a user's deletion request. Since document content is untrusted data, it could contain malicious instructions designed to manipulate the agent into performing unauthorized deletions or other unintended actions.
Ingestion points: Accessing file content via cognee.infrastructure.files.utils.open_data_file and the HTTP endpoint /api/v1/datasets/{dataset_id}/data/{data_id}/raw as described in SKILL.md.
Boundary markers: The skill explicitly instructs agents to "Confirm the exact items with the user before deleting" and to provide a preview of the candidate documents, creating a human-in-the-loop checkpoint.
Capability inventory: The skill possesses the capability to perform irreversible deletions across documents and entire datasets using cognee.forget() and cognee-cli forget.
Sanitization: No explicit sanitization or filtering of the document content is defined before the agent analyzes the preview text.
[COMMAND_EXECUTION]: The skill utilizes SDK functions and CLI commands to perform destructive data operations. While these are intended functionalities for a deletion management skill, they grant the agent the ability to execute permanent data removal commands.
Evidence: Documentation of cognee.forget(everything=True) and cognee-cli forget --everything which can wipe all datasets and invalidate session caches.

cognee-forget