The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of data from compiled binaries (untrusted external data).
Ingestion points: Data enters the agent context through SQL queries like SELECT decompile(...), SELECT ... FROM pseudocode, and SELECT ... FROM strings (mentioned in related skills).
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the agent when viewing decompiled code or strings that could contain adversarial content.
Capability inventory: The skill uses database operations (SELECT, UPDATE, INSERT) to manipulate the analysis environment. While Bash is listed in allowed-tools, no shell scripts or network operations are present in the skill files.
Sanitization: The workflow does not include explicit sanitization steps for external content before it is processed by the agent; it relies on the underlying platform's safety guardrails.

re-source