claude-api

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: References official and trusted Anthropic SDKs (anthropic and @anthropic-ai/sdk) for installation via standard package managers.
  • [COMMAND_EXECUTION]: Includes standard shell commands for environment configuration and package management, which are typical for the skill's purpose.
  • [CREDENTIALS_UNSAFE]: Correctly advises against hardcoding API keys and provides generic placeholders for environment setup instructions.
  • [DATA_EXFILTRATION]: Demonstrates reading a local image file for the Claude Vision API using standard Python patterns, consistent with documented API features.
  • [PROMPT_INJECTION]: The skill provides templates for processing user messages. Evidence chain: (1) Ingestion points: user input fields in message payloads; (2) Boundary markers: absent; (3) Capability inventory: API network operations, tool execution loops, and local file reading; (4) Sanitization: absent. While this establishes a surface for indirect prompt injection, it represents standard API integration patterns and is consistent with the primary purpose of the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 12:31 AM