The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches external content from the web during the research phase to ground its articles. This content is used as context for the 'TopicExpert' persona to answer questions and for the final article generation.
Ingestion points: Web search results and fetched page content are processed in the research phase (references/research.md) and stored in research/conversations.jsonl and sources.json.
Boundary markers: The skill does not explicitly instruct the agent to use delimiters or 'ignore' instructions embedded within the retrieved snippets, although it does apply a strip_citations function to prevent citation confusion.
Capability inventory: The skill has significant file-write capabilities across the research/, outline/, and article/ directories, and uses network tools for fetching.
Sanitization: Sanitization is focused on citation hygiene (strip_citations) rather than filtering potentially malicious instructions or scripts embedded in fetched Markdown or HTML content.
[DATA_EXFILTRATION]: The skill supports a --docs <dir> argument which allows the agent to ingest a local directory as a source pool for research.
Evidence: SKILL.md and references/research.md describe the use of local files provided via the --docs parameter.
Risk: While this is a primary feature, there is a risk that a user (or an indirect injection) could point the agent toward sensitive directories (e.g., ~/.ssh or ~/.aws). The agent might then include sensitive data from these files in its research 'snippets' or 'answers', effectively exposing them in the final article or the research/ artifacts.

storm