The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted markdown data through scripts/analyze_file.py and scripts/communication_runtime/markdown_source.py. This data is used to generate communication analysis reports and is optionally refined using an LLM via the OpenAI API in scripts/communication_runtime/llm.py. The refined content is then written to the local file system in scripts/communication_runtime/render.py. The lack of explicit boundary markers or sanitization for the input text creates a surface for indirect prompt injection where malicious content in a processed file could influence the final report or downstream tasks that rely on the generated JSON artifact.
Ingestion points: The load_markdown_source function in scripts/communication_runtime/markdown_source.py reads content from user-provided markdown files.
Boundary markers: The skill does not use specific delimiters or instructions to prevent the LLM from following instructions embedded within the analyzed transcripts or notes.
Capability inventory: The skill performs file system writes in scripts/communication_runtime/render.py to save generated reports in both Markdown and JSON formats.
Sanitization: There is no evidence of escaping, validation, or filtering of the external markdown content before it is processed by the analysis engine or sent to the LLM for refinement.

communication-analysis