The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates communication analysis reports from a user-provided directory, creating a vulnerability surface where malicious instructions could be embedded in the source artifacts.\n- Ingestion points: The _discover_reports function in scripts/summarize_folder.py recursively reads and parses all JSON files found in analysis/ subdirectories throughout the provided folder tree.\n- Boundary markers: The output generated by render_summary_markdown and write_summary in scripts/summarize_folder.py does not utilize delimiters (like XML tags or triple backticks with specific headers) or include explicit system-level instructions to the agent to disregard potential commands contained within the summarized content.\n- Capability inventory: The script performs file reads and writes. The resulting summary files (analysis/index.md and analysis/index.json) are intended to be read by the AI agent, which may then perform further privileged actions based on the summarized data.\n- Sanitization: The script performs no validation or sanitization on the text fields extracted from the JSON reports (such as label, why_it_matters, or evidence_text) before injecting them directly into the markdown and JSON output files.

communication-summary