The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill describes loading external shared libraries (.so, .dylib, .dll) at runtime via conn.load_extension() and registering Python callables as SQL functions that execute within the database query engine.
[COMMAND_EXECUTION]: The documentation includes platform-specific build commands using gcc, cl, and clang to compile C source code into executable shared libraries.
[INDIRECT_PROMPT_INJECTION]: The skill identifies an attack surface where a malicious database schema (triggers, views, generated columns) could invoke registered functions. The following evidence chain is identified:
Ingestion points: Database files and schemas processed by the agent's SQLite engine (SKILL.md).
Boundary markers: Instructions provided to use PRAGMA trusted_schema = OFF to block untrusted schema execution (SKILL.md).
Capability inventory: Execution of Python UDFs and loading of binary shared libraries (SKILL.md).
Sanitization: Discusses using DIRECTONLY and INNOCUOUS flags to control function exposure in schema-defined SQL (SKILL.md).

sqlite