The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted GraphQL schemas and database metadata to generate code, READMEs, and agent-specific documentation. This creates a vulnerability where a malicious schema could embed instructions to manipulate the agent during generation or during subsequent consumption of the generated files.
Ingestion points: Remote GraphQL introspection endpoints, local .graphql files, and database metadata accessed via connection strings.
Boundary markers: The skill does not provide explicit markers or instructions to isolate schema-derived text from the agent's logic.
Capability inventory: The skill is capable of writing generated files to the filesystem and executing the @constructive-io/graphql-codegen CLI.
Sanitization: The instructions do not mention sanitizing or escaping schema-provided strings used in the output artifacts.
[COMMAND_EXECUTION]: The documentation instructs the agent to run the @constructive-io/graphql-codegen package via npx and pnpm. This execution of external code is considered safe as it originates from the skill's verified author, 'constructive-io'.
[EXTERNAL_DOWNLOADS]: The skill connects to user-defined GraphQL endpoints and database servers to perform schema introspection. These network operations are intrinsic to the tool's function and target resources explicitly provided by the user.

constructive-codegen