The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection because it processes untrusted content from multiple external sources to generate research briefs.
Ingestion points: Fetches data from general URLs (WebFetch), search engine results (WebSearch), and social media platforms including Reddit, X, YouTube, and Hacker News via the last30days tool.
Boundary markers: Absent. There are no instructions or delimiters provided to ensure the agent ignores malicious directives that might be hidden in the fetched web content.
Capability inventory: The agent can write files to the local configuration directory (~/.config/makerskills/), query the Notion API using a provided $NOTION_API_KEY, and execute shell commands (grep, curl).
Sanitization: Absent. The skill does not specify any sanitization or filtering of external data before it is passed to the synthesis step.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell-level commands for data discovery and archive management.
Uses grep to search through local memory files and previous research archives stored in the user's home directory (~/.config/makerskills/).
Uses curl in conjunction with defuddle to fetch web content when standard agent tools are unavailable.
Relies on the agent-browser CLI tool to interact with complex or authenticated web pages.

deep-research