The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute scripts directly in the shell using patterns like curl ... | bash and irm ... | iex. This allows for the execution of arbitrary code from a remote source without prior verification.
Evidence in SKILL.md: curl -fsSL https://raw.githubusercontent.com/d4rkNinja/infynon-cli/main/scripts/install.sh | bash
Evidence in SKILL.md: irm https://raw.githubusercontent.com/d4rkNinja/infynon-cli/main/scripts/install.ps1 | iex
[EXTERNAL_DOWNLOADS]: The skill references external resources for installation hosted on the author's GitHub repository (d4rkNinja/infynon-cli). While GitHub is a well-known service, the specific repository is an individual user account and not a pre-verified trusted organization.
Evidence: Downloads from raw.githubusercontent.com and github.com releases.
[COMMAND_EXECUTION]: The skill requires the Bash tool to perform installations and execute package management commands through a CLI wrapper. This gives the skill the capability to execute system-level commands and interact with various package managers.
Evidence: Extensive use of infynon pkg <command> which invokes underlying tools like npm, pip, and cargo via shell execution.
[PROMPT_INJECTION]: The skill incorporates potential surfaces for indirect prompt injection by instructing the agent to process structured JSON output from the CLI tool. If the tool processes malicious metadata from untrusted packages, it could influence downstream agent behavior.
Ingestion points: SKILL.md (instructions to parse --json output for automated workflows).
Boundary markers: Absent.
Capability inventory: Bash tool with access to package managers and file system writes.
Sanitization: Absent.

package-security