The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a callback mechanism to capture and store verified identity attributes (names, emails, SSO domains) from external identity providers. The provided code examples store these values directly into the application's persistent state (e.g., emails.add(caller, email)) without showing sanitization or validation steps. This creates a potential surface for indirect injection if these attributes are subsequently used in contexts that interpret them as instructions.
Ingestion points: The attrs parameter in the MixinAuthorization callback within SKILL.md (e.g., attrs.email, attrs.name).
Boundary markers: None are present in the documentation examples to delimit the ingested text from internal prompts.
Capability inventory: The skill manages internal state updates. No subprocess execution, arbitrary file writing, or network exfiltration operations are directly triggered by this ingested data in the snippets.
Sanitization: The documentation examples omit sanitization or length-validation logic for the externally sourced attribute strings.
[COMMAND_EXECUTION]: The skill and its migration guide reference the installation of development dependencies using mops and npm. These packages (caffeineai-authorization, @caffeineai/core-infrastructure) are official resources provided by the skill's author (caffeinelabs) for implementing core platform functionality.

extension-authorization