The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains specific hardcoded example passwords within Python and shell code blocks which may be inadvertently executed or used.
Evidence in SKILL.md: admin_password="SecureP@ss123!" and password="WalletP@ss456!".
[DATA_EXPOSURE]: The skill accesses the local OCI configuration file, which typically contains sensitive credentials such as API keys and fingerprints.
Evidence in SKILL.md: oci.config.from_file("~/.oci/config").
[INDIRECT_PROMPT_INJECTION]: The skill retrieves and prints metadata (table names, object statuses) from an external database, creating a surface for injection if an attacker controls the database schema.
Ingestion points: Reads from all_tables and all_objects metadata tables in SKILL.md.
Boundary markers: None present to delimit untrusted database content.
Capability inventory: Shell command execution (expdp), local file system writes (zipfile.extractall), and OCI SDK network operations.
Sanitization: No sanitization or validation is performed on the data retrieved from the database queries.
[EXTERNAL_DOWNLOADS]: The skill installs official Oracle Cloud and database driver packages.
Evidence in SKILL.md: pip install oci and pip install oracledb.
[COMMAND_EXECUTION]: The skill executes system commands for database exporting and OCI resource listing.
Evidence in SKILL.md: Usage of expdp and oci db autonomous-database list commands.

oraclecloud-schema-migration