The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process text from user-provided PowerPoint presentations to perform its analysis. Because the agent reads every slide, there is a possibility that a slide could contain hidden instructions or malicious prompts intended to influence the agent's behavior or bias its report.
Evidence: The instructions in SKILL.md direct the agent to "Pull text from every slide" and "Read every slide, then report findings."
Context: This is an inherent risk when an agent processes untrusted external data. The use of a structured report template in references/report-format.md helps to ground the agent's output.
Local Command Execution: The skill invokes a bundled Python script to perform numerical consistency checks. This involves the agent executing a shell command to process a temporary file containing the presentation content.
Evidence: SKILL.md contains the instruction: python scripts/extract_numbers.py /tmp/deck_content.md --check.
Context: The script scripts/extract_numbers.py uses standard Python libraries for regex-based extraction and does not perform network operations or access sensitive system paths beyond the input file provided to it.

ib-check-deck