The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Bible data from an external API (FHL). It mitigates injection risks by using clear boundary markers (<<>> fences) and explicit instructions to the AI agent to ignore any commands found within the data. These measures are found in server.py and codex_backend.py.
[COMMAND_EXECUTION]: The skill uses uv run and the codex CLI via asyncio.create_subprocess_exec to manage its environment and run the backend. These executions are scoped to the skill's operational requirements and, in the case of Codex, specifically use a read-only sandbox with shell tools and web search disabled for security.
[DATA_EXPOSURE]: The web server binds strictly to the loopback interface (127.0.0.1), ensuring that the local server is not accessible from the network. User notes are stored locally on the user's Desktop in a dedicated directory, with no evidence of unauthorized data transmission.
[EXTERNAL_DOWNLOADS]: The skill fetches theological data from a well-known academic and religious service (bible.fhl.net). Frontend dependencies are either vendored locally or loaded from reputable CDNs using Subresource Integrity (SRI) hashes to prevent supply chain attacks.
[OBFUSCATION]: Analysis of the source code and vendored JavaScript libraries found no signs of malicious obfuscation, encoded payloads, or hidden commands.

bible-buddy-web