claude-code-guide
Pass
Audited by Gen Agent Trust Hub on Aug 8, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill's documentation suggests using phrases like "forget previous instructions" to troubleshoot agent behavior. While intended for debugging, this pattern is used in prompt injection attacks to bypass system constraints.
- Evidence: "ask the agent to 'forget previous instructions' if confused" in SKILL.md.
- [PROMPT_INJECTION]: The skill defines a workflow where an agent reads project files and executes commands based on guidelines in CLAUDE.md, which creates an indirect prompt injection surface.
- Ingestion points: Project files (README.md, file content) read by the agent.
- Boundary markers: Absent; the guidelines do not specify delimiters for external content.
- Capability inventory: Shell command execution (npm run, npm test, npm build) and file system access.
- Sanitization: Absent; the skill does not instruct the agent to sanitize or validate content from project files before processing.
Audit Metadata