The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads potentially untrusted external data from the codebase, including database schemas, migration files, and query logs (identified in SKILL.md under the ACQUIRE scene and resources/execution-protocol.md Step 1). It combines this data ingestion with the capability to write new files and execute local commands, creating a surface for indirect prompt injection. The documentation does not specify boundary markers or input sanitization logic to separate this untrusted content from the agent's instructions.
[DYNAMIC_EXECUTION]: The skill generates and executes database-related content at runtime, specifically schema migrations and query plans (defined in SKILL.md and resources/migration-playbook.md). This process involves writing DDL/DML statements to the filesystem and then executing them via local toolchains, which is a standard but sensitive operation for database management.
[COMMAND_EXECUTION]: The skill is configured to execute local system commands to perform its core functions, such as running project-specific migration tools and query plan analyzers (EXPLAIN ANALYZE). This functionality is documented in the SKILL.md canonical workflow and resources/query-tuning.md, representing a direct use of system resources to interact with the database environment.

oma-db