The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install software by downloading a script from a remote URL and piping it directly into the bash shell. This is a highly dangerous practice as it allows for unverified code to be executed on the user's system.
Evidence: curl -s https://install.ladybugdb.com | bash in SKILL.md and references/cli.md.
[COMMAND_EXECUTION]: The skill provides numerous commands for interacting with the local file system, managing databases, and performing network operations (via ATTACH). While these are functional requirements for a database tool, they significantly expand the attack surface if the agent is manipulated into executing them with malicious intent.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a large attack surface for indirect prompt injection due to its core functionality of ingesting and processing untrusted data from multiple external sources.
Ingestion points: The COPY FROM, LOAD FROM, and ATTACH commands (found in references/import.md and references/attach.md) allow the database to read from CSV, Parquet, JSON, and external databases (PostgreSQL, Neo4j, etc.).
Boundary markers: No specific boundary markers or instructions to ignore embedded commands within the processed data were observed in the Cypher query examples.
Capability inventory: The skill has access to local file writing (COPY TO), network communication (ATTACH), and complex data processing capabilities through graph algorithms and LLM integrations.
Sanitization: There is no mention of sanitization, validation, or filtering mechanisms for the data being imported from these external sources.

ladybugdb