The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and process content from IDA Pro databases, such as assembly comments and decompiled pseudocode, which may contain adversarial instructions. \n
Ingestion points: Untrusted data enters the agent's context through queries to the pseudocode, comments, funcs, and bookmarks tables as seen in SKILL.md and references/annotation-workflows.md. \n
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when handling strings retrieved from the database to prevent the agent from obeying embedded commands. \n
Capability inventory: The skill utilizes the Bash tool and has the ability to modify the IDA database through UPDATE, INSERT, and DELETE SQL operations. \n
Sanitization: There is no evidence of string sanitization, escaping, or validation performed on external content retrieved from the database before it is used by the agent.

annotations