The Agent Skills Directory

[METADATA_POISONING]: The skill's YAML frontmatter identifies 'prisma' as the author, which conflicts with the actual provider information. This deceptive attribution may lead users to mistakenly assume the skill has official provenance and inherent safety.
[INDIRECT_PROMPT_INJECTION]: The skill documents commands that incorporate external or agent-generated strings into privileged execution contexts.
Ingestion points: The prisma init --prompt command takes natural language descriptions for schema generation, and prisma db execute allows for the execution of arbitrary SQL from files or stdin.
Boundary markers: No delimiters or 'ignore embedded instructions' warnings are defined for the input passed to these CLI flags.
Capability inventory: The Prisma CLI has broad capabilities to modify database schemas, perform raw SQL operations (including data deletion), and write project configuration files.
Sanitization: There is no mention of sanitization or validation logic for strings passed into the prompt-based generation or SQL execution features.
[COMMAND_EXECUTION]: The skill details numerous commands that perform sensitive database operations, such as prisma db push --force-reset and prisma migrate reset, which are capable of full data destruction if misused.
[DYNAMIC_EXECUTION]: The documentation for prisma db seed encourages the runtime execution of user-defined TypeScript or JavaScript scripts (e.g., tsx prisma/seed.ts), which constitutes a vector for the execution of dynamic local assets.

prisma-cli