The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses persona adoption, instructing the agent to act as a 'first-century Jewish Torah scholar' and providing a detailed hermeneutic framework that overrides standard theological interpretations with a specific historical-critical perspective.
[EXTERNAL_DOWNLOADS]: Fetches scripture and historical data from multiple external sources including sefaria.org, biblegateway.com, chinesebible.org.hk, pseudepigrapha.com, sacredbible.org, and github.com. It also downloads browser binaries (Chromium) during its setup phase to facilitate web scraping.
[COMMAND_EXECUTION]: Executes shell commands via Python's subprocess module to manage data repositories (e.g., cloning the LXX dataset from GitHub) and uses browser automation via the patchright library to extract text from websites.
[PROMPT_INJECTION] (Indirect): The skill ingests data from various third-party websites into the agent's context. While these are established religious archives, they represent a surface for indirect prompt injection if a source were to be compromised.
Ingestion points: Content is fetched from multiple external domains via scripts like scripts/fetch_sefaria.py and scripts/fetch_url.py.
Boundary markers: The skill does not explicitly use delimiters or instructions to ignore embedded commands within the fetched scripture text.
Capability inventory: The skill has access to powerful tools including Bash, Write, and Edit. Some bundled scripts also utilize subprocess for system operations.
Sanitization: While the scripts clean HTML tags, there is no verification or sanitization to detect potentially malicious instructions embedded in the historical or religious texts.

bible-buddy