ask-claude

Pass

Audited by Gen Agent Trust Hub on Sep 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands (such as claude -p, cd, and mktemp) to interact with the environment and execute the Claude Code CLI in non-interactive mode.
  • [EXTERNAL_DOWNLOADS]: The skill documentation references official resources and documentation hosted on Anthropic's domains (anthropic.com, claude.com), which are recognized as trusted services for AI-related tools.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes local repository content by passing it as context to the Claude Code CLI. It mitigates potential risks by providing structured prompt templates that use XML tags (e.g., <context>, <task>) as boundary markers and recommending the use of the CLI's --permission-mode plan (read-only) to restrict the scope of external tool actions.
  • Ingestion points: Local repository files are read when executing the CLI within a target directory (e.g., cd <DIR>).
  • Boundary markers: Recommends the use of XML tags to clearly separate data context from instructions.
  • Capability inventory: Limited to shell subprocess execution for interacting with the CLI.
  • Sanitization: Relies on the pre-installed CLI's permission modes and the agent's internal path handling.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 24, 2026, 09:40 AM