The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill's prerequisite installation command (curl -LsSf https://astral.sh/uv/install.sh | sh) uses the 'pipe to shell' pattern. This is a high-severity risk as it executes an unverified remote script with the user's local permissions, bypassing package integrity checks. Although intended for the 'uv' tool, the source is not on the trusted whitelist provided in the security skill scope.- EXTERNAL_DOWNLOADS (LOW): The skill performs multiple external network operations to non-whitelisted domains, including fetching competition content via the GitHub API and downloading datasets via the crunch download command. While functional for the skill's purpose, these are unverified external dependencies.- CREDENTIALS_UNSAFE (LOW): The documentation in references/competition-setup.md explicitly suggests storing competition tokens in plaintext at ~/.crunch/.tokens. This practice exposes sensitive credentials to any other process or AI agent with filesystem access.- PROMPT_INJECTION (LOW): The skill analyzes external code and documentation fetched at runtime from GitHub, which constitutes an indirect prompt injection surface. An attacker controlling the remote repository could embed instructions to manipulate the agent's analysis or suggestions. Evidence Chain: 1. Ingestion points: GitHub API content from crunchdao repositories; 2. Boundary markers: Absent; 3. Capability inventory: Local testing execution (crunch test), package installation (uv pip), and submission (crunch push); 4. Sanitization: Absent.

crunch-compete