The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interpolates user-supplied research topics directly into shell commands.
Evidence: The fallback mode executes gemini --yolo "[research prompt]" where the prompt is derived from the user's research topic. This allows an attacker to execute arbitrary shell commands by including shell metacharacters (e.g., ;, |, `) in the research topic.
Evidence: The primary mode uses curl to send user data to a local API: curl -s -X POST http://localhost:8005/api/search -d '{"query": "[research query]", "mode": "research"}'. Improper escaping of the [research query] variable could lead to JSON or shell injection.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to a local search broker.
Evidence: Multiple curl commands interact with http://localhost:8005/api/search and http://localhost:8005/api/extract to perform research and content extraction.
[PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection (Category 8).
Ingestion points: The [topic] parameter in the /research command (SKILL.md).
Boundary markers: Absent. User input is interpolated directly into strings without delimiters or sanitization instructions.
Capability inventory: The skill has the ability to execute shell commands via curl, python, and the gemini CLI tool.
Sanitization: No evidence of input validation, escaping, or filtering for shell-sensitive characters.

research