The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill promotes installation via bash <(curl ...) in its README.md and scripts/one_line_install.sh. This pattern downloads and executes code from the author's repository without integrity verification.
[COMMAND_EXECUTION]: Several scripts (install_from_github.sh, setup.sh, manage_evolver.sh) perform extensive shell operations. The installation scripts execute apt-get commands, modify system-wide paths like /usr/local/bin/, and manipulate the user's ~/.ssh/config file, which constitutes privilege escalation.
[EXTERNAL_DOWNLOADS]: The installation process fetches Node.js configuration from deb.nodesource.com (a well-known service) and clones external repositories from GitHub (keoy7am/skill-evomap and EvoMap/evolver) to set up its components.
[PROMPT_INJECTION]: The skill has a significant surface for indirect prompt injection (Category 8).
Ingestion points: evomap_client.py fetches "promoted assets" (Genes and Capsules) from the evomap.ai marketplace.
Boundary markers: No explicit boundary markers or "ignore embedded instructions" warnings are implemented when processing these remote assets.
Capability inventory: The skill manages a Node.js process via scripts/manage_evolver.sh and provides a Python client capable of interacting with the local workspace and remote network.
Sanitization: There is no evidence of sanitization or strict schema validation for the fetched natural language "Genes" or implementation "Capsules" before they are integrated into the agent's workflow.

evomap