The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the capability to ingest data from external sources, which constitutes a potential attack surface for indirect prompt injection if malicious instructions are embedded in the source data.
Ingestion points: Data enters the agent's context through imports from external memory systems (Mem0, Zep/Graphiti, Letta, LangMem) and COGX portable archives via the cognee.remember() and COGXArchiveSource interfaces as described in SKILL.md.
Boundary markers: The instructions do not specify the use of explicit delimiters or specific guardrail prompts to isolate untrusted external content during the import process.
Capability inventory: The skill documents capabilities for file system writes (cognee.export), database schema and data modifications (cognee-cli upgrade/downgrade), and network operations (cognee.push to Cognee Cloud).
Sanitization: The skill identifies a mode="re-derive" for data imports which involves LLM extraction from raw content, potentially allowing instructions in the raw source to influence the agent's extraction behavior.
[DATA_EXFILTRATION]: The skill describes the cognee.push() command and cognee-cli push, which exports dataset archives to Cognee Cloud. This is a documented administrative feature for the cognee platform.
[CREDENTIALS_UNSAFE]: The documentation references the use of environment variables for database credentials (e.g., MIGRATION_DB_PASSWORD, MIGRATION_DB_USERNAME), which is a standard security practice. Additionally, it warns that using include_permissions=True during export will include password hashes in the archive and explicitly instructs the user to treat those archives as secrets.

cognee-migrations