The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill offers to install missing CLI tools (reddit-cli and brave-cli) by downloading scripts from a remote GitHub repository and piping them directly into shell interpreters (bash and sh). This is a dangerous execution pattern that runs unverified remote code with the user's local privileges.
Evidence: curl -fsSL https://raw.githubusercontent.com/gupsammy/reddit-cli/main/install.sh | bash and curl -fsSL https://raw.githubusercontent.com/gupsammy/brave-cli/main/install.sh | sh in SKILL.md.
[COMMAND_EXECUTION]: User-supplied input (the research topic) is directly interpolated into multiple shell commands for tools like reddit-cli, bird, and brave-cli. Without proper sanitization, a user-provided topic containing shell metacharacters could lead to arbitrary command execution on the host system.
Evidence: reddit-cli search "{TOPIC}", bird search "{TOPIC} ...", and brave-cli search "{TOPIC}" in SKILL.md.
[DATA_EXFILTRATION]: The skill accesses the sensitive file path ~/.secrets to check for the presence of a BRAVE_API_KEY. Accessing or reading from files used to store environment secrets and credentials is a high-risk activity that could lead to the exposure of sensitive data.
Evidence: grep -q "BRAVE_API_KEY" ~/.secrets 2>/dev/null in SKILL.md.
[PROMPT_INJECTION]: The skill processes untrusted external data from Reddit, X (Twitter), YouTube transcripts, and various web sources. It lacks boundary markers or specific instructions to prevent the agent from obeying malicious instructions embedded within this third-party content, making it vulnerable to indirect prompt injection.
Ingestion points: Results from reddit-cli, bird, yt_research.py transcripts, and WebSearch in SKILL.md.
Boundary markers: Absent throughout the instruction set.
Capability inventory: The skill has extensive capabilities including bash command execution, Python script execution, and network operations.
Sanitization: No sanitization of ingested external content is performed before processing.

run-research