The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download source code from an unverified personal GitHub repository rather than an official vendor organization.
Evidence: git clone https://github.com/rak7777/mythic-mcp-proxy.git in SKILL.md.
[REMOTE_CODE_EXECUTION]: The installation process involves executing setup scripts and running server code from the unverified external repository.
Evidence: Commands pip install -r requirements.txt and python -m logos.server follow the untrusted download.
[COMMAND_EXECUTION]: The skill requires the execution of shell commands to set up and run the router service.
Evidence: Shell blocks in SKILL.md describing installation and server startup.
[INDIRECT_PROMPT_INJECTION]: The skill's primary functionality involves processing untrusted external data such as research papers and code patches, which can contain hidden instructions to manipulate the model's reasoning.
Ingestion points: task.load_document("research_paper.pdf") and router.query(prompt=f"Review this code change:\n\n{diff}") in SKILL.md.
Boundary markers: None present; the skill interpolates raw external content directly into reasoning prompts.
Capability inventory: The skill performs network operations (HTTP/WebSockets) and local file reads.
Sanitization: No sanitization or validation of external input is demonstrated.

logos-router-reasoning