skills/xiaolai/vmark/ai-coding-agents/Gen Agent Trust Hub

ai-coding-agents

Pass

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: SAFEPERSISTENCEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill includes instructions to add shell completions to profile files like .bashrc and .zshrc. This is a standard and user-initiated procedure for enabling CLI auto-completion.
  • [EXTERNAL_DOWNLOADS]: Installation commands are provided for official packages @openai/codex and @anthropic-ai/claude-code from standard package registries. These resources belong to well-known and trusted organizations.
  • [COMMAND_EXECUTION]: The skill documents various CLI commands that allow agents to perform complex tasks, such as executing shell commands, running builds, and managing project files, which is consistent with the primary purpose of coding agents.
  • [PRIVILEGE_ESCALATION]: Documentation covers safety-bypass flags provided by the tools, such as --yolo and --dangerously-skip-permissions. These are documented features of the underlying tools and are accompanied by safety warnings.
  • [INDIRECT_PROMPT_INJECTION]: The described tools ingest data from codebases and version control systems, which represents a potential attack surface.
  • Ingestion points: Source code, pull request metadata, and git diffs (SKILL.md, references/claude-code-advanced.md).
  • Boundary markers: Guidance on using instructions files such as CLAUDE.md and project settings to define agent constraints.
  • Capability inventory: Execution of arbitrary shell commands, file system modification, and network access via MCP servers.
  • Sanitization: The documentation recommends review checkpoints for diffs and the use of budget or turn limits to maintain control over agent operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 7, 2026, 04:33 PM