The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's 'Agentic Protocol' defined in SKILL.md requires the use of tools like web search and fetch_webpage to gather real-world economic data and statistical information.
[COMMAND_EXECUTION]: The push-to-github.bat script executes standard Git commands (git push) and uses the start command to open a legitimate GitHub repository creation URL in the user's default browser. Additionally, SKILL.md mentions using run_in_terminal to acquire statistical data during the research phase.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external websites via fetch_webpage and web search. Evidence in SKILL.md shows the agent is instructed to process this data to reveal 'contradictions' and 'class dynamics' without explicit boundary markers or sanitization steps to isolate the external content from the agent's core instructions.
[DATA_EXFILTRATION]: While the skill accesses the internet for research, there are no patterns suggesting the exfiltration of sensitive local data. The research results are stored locally in the references/research/ directory as specified in SKILL.md.

marx-perspective