The Agent Skills Directory

[SAFE]: The skill instructions focus on secure configuration of database access. It promotes the use of Oracle proxy authentication to create a 'logon-only' agent user, which is a significant security improvement over shared credentials. It also provides explicit instructions to protect sensitive files like connections.json using .gitignore and validation checks via pythia doctor to ensure credentials are not tracked by version control.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with external database content, creating an inherent attack surface for indirect prompt injection.
Ingestion points: Database object metadata, names, and code read via pythia check, pythia similar, and pythia plscope (reference/environment.md).
Boundary markers: The instructions highlight that Oracle grants and proxy authentication are the primary layers of protection that cannot be bypassed by the agent.
Capability inventory: The toolset includes capabilities for writing to the database (pythia apply), creating users (pythia agent-user), and modifying agent configuration files (pythia install in SKILL.md).
Sanitization: The skill recommends using pythia unistr for string literals to ensure source survival and likely handles basic escaping, though explicit sanitization of database-returned strings is handled by the underlying toolset.
[COMMAND_EXECUTION]: The skill involves standard environment setup commands such as pip install pythia-plsql and docker run for database containers. These use well-known, official sources (container-registry.oracle.com) and standard package managers.

pythia-setup