The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a local file to establish its operational context, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: The skill is instructed to locate and extract venture data, customer profiles, and assumptions from a file named innov.local.md in the working directory.
Boundary markers: The instructions lack specific delimiters or "ignore embedded instructions" warnings for the agent when reading the contents of the external markdown file.
Capability inventory: The analysis did not detect any high-risk capabilities such as network exfiltration, arbitrary command execution, or file writing tools that could be abused via an injection.
Sanitization: There are no explicit validation or sanitization steps defined for the data extracted from the local context file.

canvas