The Agent Skills Directory

[COMMAND_EXECUTION]: The script research-workflow.js dynamically constructs shell commands (grep and find) by interpolating participant names directly into command strings. Since these names are fetched from an external source (Cal.com), a malicious participant name containing shell metacharacters (e.g., ;, `, $(...)) could lead to arbitrary command execution on the host system.- [DATA_EXFILTRATION]: The skill accesses and reads private meeting notes and person profiles from the user's Obsidian vault. This sensitive information is then passed to LLM subagents for research and summarization, effectively exposing private local data to the model provider.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through external data sources.
Ingestion points: Participant names, email domains, and event descriptions from Cal.com, along with content from web search results, are processed by LLM agents.
Boundary markers: No clear delimiters or safety instructions are used in research-workflow.js to separate untrusted external data from the agent's core instructions.
Capability inventory: The skill possesses the ability to execute shell commands (grep, find) and write files to the local file system.
Sanitization: There is no evidence of input validation, escaping, or filtering of the ingested data before it is used in prompts or system commands.

meeting-prep