The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill exhibits a significant attack surface for indirect prompt injection by instructing the agent to ingest and summarize data from internal communication platforms where content may be controlled or influenced by malicious actors.
Ingestion points: Data is ingested from Slack, Google Drive, Email, and Calendar as specified in examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md.
Boundary markers: Absent. The skill does not provide delimiters or instructions to the agent to disregard instructions embedded within the retrieved data.
Capability inventory: The agent utilizes tool-based read access to aggregate information for text generation.
Sanitization: Absent. There is no mention of filtering or validating the content retrieved from external tools before processing.
Data Exposure Surface (SAFE): The skill is designed to process sensitive internal data (emails, docs, Slack). While it does not contain exfiltration code (e.g., curl/wget to external domains), the aggregation of this data into summaries increases the potential impact if the agent's output is subsequently leaked or misdirected.

internal-comms