The Agent Skills Directory

[DATA_EXPOSURE]: The skill exposes multiple endpoints designed to harvest sensitive credentials directly from the local filesystem. Specifically, POST /api/providers/agy-auth/apply-local, POST /api/providers/{id}/claude-auth/apply-local, and POST /api/providers/{id}/codex-auth/apply-local are documented to 'auto-detect and import local login' tokens from disk, targeting sensitive configuration files for CLI tools and browser-based sessions.
[DATA_EXPOSURE]: The API includes explicit credential extraction capabilities via endpoints such as POST /api/providers/{id}/claude-auth/export and POST /api/providers/{id}/codex-auth/export, which allow for the retrieval and exfiltration of stored authentication data.
[COMMAND_EXECUTION]: The endpoint documentation for /api/providers/validate, /api/providers/bulk, and /api/providers/import explicitly references a 'devin CLI-spawn fallback' mechanism. This indicates the underlying server implementation executes shell commands to process provider validation and bulk operations, which can be a vector for command injection or unauthorized system access.
[INDIRECT_PROMPT_INJECTION]: The POST /api/provider-models/validate-and-add workflow creates a vulnerability surface by processing live inference responses from external, untrusted AI providers.
Ingestion points: External provider responses during 'streamed synthetic tool-call and continuation proofs'.
Boundary markers: None documented; the process is described as an atomic read-back after proof passage.
Capability inventory: The skill has full CRUD access to the provider catalog and the ability to trigger network requests for validation.
Sanitization: Mentions 'strict support' and 'bounded generation', but these do not prevent adversarial instructions embedded in the provider's response from influencing the agent's state or the resulting configuration.

omni-providers