skill-manager

Fail

Audited by Gen Agent Trust Hub on Nov 2, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to download a setup script from the developer's GitHub repository (github.com/Melvynx) and execute it using bash.
  • [EXTERNAL_DOWNLOADS]: The skills.sh script utilizes curl to fetch skill archives from GitHub (codeload.github.com) and can also utilize the GitHub CLI (gh api) to access content from private repositories.
  • [COMMAND_EXECUTION]: The skill performs various system operations including file copying (cp -R), directory creation (mkdir -p), and symlinking (ln -sfn). It also supports executing commands on remote servers via SSH using the bash -s pattern to pipe the management script to the remote shell.
  • [INDIRECT_PROMPT_INJECTION]: The skill's architecture is designed to import instructions and scripts from external, potentially untrusted sources into the agent's core configuration.
  • Ingestion points: The install command accepts GitHub repository identifiers, URLs to skill folders, or local directory paths as sources for new skills.
  • Boundary markers: There is no evidence of content validation, integrity checking, or security scanning of the SKILL.md files or associated scripts being imported.
  • Capability inventory: Imported skills are written to active skill directories (~/.claude/skills or project-specific equivalents) where they can influence agent behavior and utilize available tools.
  • Sanitization: The installation process copies files directly from the source into the agent's skill path without auditing the instructions for malicious patterns or unauthorized capability requests.
  • [PERSISTENCE]: By installing new skills into the agent's startup paths, the skill creates persistent changes to the agent's instruction set that remain active across multiple sessions.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/Melvynx/ai-sprint-skills/main/skills/skill-manager/scripts/skills.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 2, 2026, 08:07 AM