The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and analyzes untrusted data from arbitrary URLs and user-pasted text. This creates a surface where malicious instructions embedded in the biblical content could attempt to influence the agent's behavior. \n
Ingestion points: Content is fetched from URLs via a local script or provided directly by the user as pasted text in the conversation. \n
Boundary markers: The instructions do not specify the use of clear delimiters or warnings to the AI to ignore instructions found within the audited text. \n
Capability inventory: The skill can execute shell commands and local scripts, perform network requests, and write files to the user's Desktop directory. \n
Sanitization: There is no explicit requirement for the skill to sanitize or escape the external content before it is processed by the model. \n- [COMMAND_EXECUTION]: The skill utilizes shell pipelines and executes Python scripts located in the directories of related skills belonging to the same author (bible-buddy and markdown-to-html). \n
Evidence: The skill instructs the agent to run uv run --directory {BIBLE_BUDDY} scripts/fetch_url.py "<URL>", uv run ... scripts/detect_desktop.py, and uv run ... scripts/md_to_html.py. It also employs complex shell commands such as grep '|' ... | awk -F'|' '{print $2}' | sed 's/^ *//;s/ *$//' | grep -v '^-' | grep -v '^Scripture' to process reference data. \n
Context: These resources are identified as vendor resources associated with the author. \n- [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch external content and manages tool dependencies. \n
Evidence: It uses a custom script to download text from user-provided URLs and requires a one-time setup that installs the patchright package and the Chromium browser via uv run and patchright install chromium.

bible-fact-check