The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to manage sensitive credentials including OAuth access tokens, refresh tokens, and API keys for multiple LLM providers. It instructs the agent to read these from local environment files (e.g., ~/.hermes/.env) and import them into the router. The skill mitigates risks by recommending local-only API calls (127.0.0.1) and strict file permissions (0600).
[COMMAND_EXECUTION]: The skill outlines the use of an administrative HTTP API to perform operations such as listing providers, importing credentials, and triggering token refreshes. These operations involve network requests that manipulate the underlying service configuration.
[DATA_EXFILTRATION]: The skill facilitates the movement of secrets from local files to a network endpoint (the OmniRoute API). Although this is the primary purpose of the skill and instructions focus on local communication, the capability involves reading sensitive files and transmitting their contents.
[PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection as it ingests external credential data and interpolates it into administrative API requests.
Ingestion points: Credential blobs read from host files or provided via user input (SKILL.md).
Boundary markers: Absent; JSON structures are used but no explicit prompt delimiters for the data are provided.
Capability inventory: HTTP POST requests to the Admin API for credential ingestion.
Sanitization: Absent; tokens and metadata are passed directly as structured data to the router.

omnirouter