The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the remember() function and cognee-cli, which are used to ingest external data into the agent's long-term memory and graph database. This creates a surface where malicious instructions in processed data could influence the agent's future behavior.
Ingestion points: remember() function and CLI commands referenced in SKILL.md.
Boundary markers: The instructions do not document the use of specific delimiters or safety warnings to mitigate the impact of instructions embedded within ingested data.
Capability inventory: The skill facilitates extensive capabilities including relational and vector database writes (Postgres, Neo4j, etc.), remote storage access (AWS S3), and invocation of various LLM providers.
Sanitization: No specific sanitization, validation, or filtering of the external content being "remembered" is described.
[EXTERNAL_DOWNLOADS]: The instructions refer to external community adapters for extending the system's database support.
Evidence: Directs users to download and install adapters from https://github.com/topoteretes/cognee-community.
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands to manage the environment and its state.
Evidence: Includes commands for package installation (pip install), container management (docker compose), and system maintenance via the cognee-cli (e.g., cognee-cli forget --all).
[SAFE]: The skill promotes secure configuration habits by instructing users to manage sensitive data such as LLM_API_KEY, DB_PASSWORD, and AWS credentials through .env files rather than hardcoding them in scripts.

cognee-integrations