The Agent Skills Directory

[SAFE]: Authentication and Credential Management
The skill utilizes the official Explorium OAuth2 device authorization flow and API key authentication.
Credentials (API keys and access tokens) are stored locally in the user's home directory (~/.agentsource/config.json) with restrictive file system permissions (0600), ensuring they are only accessible by the owner.
[SAFE]: Network Security
All network communications are directed exclusively to verified and expected domains: api.explorium.ai and explorium.auth0.com.
The CLI tool uses the Python standard library's urllib for requests, avoiding risks associated with external dependency vulnerabilities.
[SAFE]: Data Privacy and Context Protection
The skill implements a robust privacy model by writing large API responses to local temporary files (/tmp/agentsource_*.json). This prevents sensitive data or large payloads from entering the AI's conversation context, which protects both privacy and context window limits.
[SAFE]: Software Integrity
The installation script (setup.sh) and core logic (agentsource.py) do not download external code or scripts at runtime.
The skill is implemented using pure Python (standard library only), which eliminates supply chain risks associated with third-party packages.
[SAFE]: Indirect Prompt Injection Mitigation
The skill includes specific instructions to the agent to only ingest metadata and small samples from user-provided CSV files, reducing the surface area for potential indirect prompt injection attacks from untrusted data sources.

vibe-prospecting