manual-interacting-with-claude-code-via-cli
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions for executing shell commands via the
claudeCLI. This includes commands for environment setup, diagnostic checks, and operational workflows. - [COMMAND_EXECUTION]: The instructions explicitly mention the use of high-risk flags such as
--dangerously-skip-permissionsandbypassPermissions. While the skill mandates their use only within isolated environments like containers or VMs, these flags facilitate the removal of tool-call and file-access restrictions within the CLI. - [EXTERNAL_DOWNLOADS]: The skill includes multiple reference links to official documentation on the
code.claude.comdomain. These are links to a well-known service provided for instructional context and are considered safe under standard usage. - [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface (Category 8) by instructing the agent to ingest and analyze output from CLI commands and subagents.
- Ingestion points: Data enters the context from
claude -poutputs and verbose execution streams (documented inSKILL.md). - Boundary markers: The instructions lack specific delimiters or warnings to ignore adversarial instructions embedded in the ingested CLI output.
- Capability inventory: The skill uses
bashfor direct command execution across multiple scenarios. - Sanitization: There are no provided instructions for sanitizing, escaping, or validating the output from the CLI before it is processed by the agent.
Audit Metadata