The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes data from an external local file (kg.csv) and presents it to the agent without sanitization or protective boundary markers.
Ingestion points: Data is ingested in scripts/query_primekg.py via the pd.read_csv function which loads the entire knowledge graph.
Boundary markers: The skill does not use delimiters or instructions to the agent to treat the retrieved biological data as untrusted or to ignore any potential instructions embedded within the text fields.
Capability inventory: The associated scripts use standard data processing libraries (pandas); no dangerous capabilities like shell execution (subprocess), dynamic evaluation (eval/exec), or network operations were identified.
Sanitization: There is no evidence of content validation, escaping, or filtering of the CSV data before it is returned to the agent's context.
[SAFE]: The skill uses hardcoded absolute paths pointing to a specific user's directory (/mnt/c/Users/eamon/Documents/Data/PrimeKG/kg.csv). While this discloses the local system's directory structure and a username, it does not target sensitive files such as credentials, private keys, or configuration secrets.

primekg