The Agent Skills Directory

[SAFE]: No malicious patterns detected across all evaluated categories. The skill's implementation follows security best practices for its intended purpose.\n- [UNVERIFIABLE_DEPENDENCIES]: The skill requires the 'anthropic' Python package, the official client library for the Anthropic API. This is a well-known, trusted dependency.\n- [DATA_EXPOSURE]: The skill demonstrates secure management of sensitive API keys by retrieving them from environment variables or project-specific configuration files (e.g., '/mnt/project/ANTHROPIC_API_KEY.txt'). No hardcoded credentials or unauthorized data transmission were found.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides an interface for processing untrusted data via sub-agent calls. 1. Ingestion points: 'prompt' and 'system' parameters in orchestration functions (scripts/claude_client.py). 2. Boundary markers: Structured API roles (system, user, assistant) are used to differentiate content. 3. Capability inventory: The skill is capable of network operations to the Anthropic API and reading local configuration files. 4. Sanitization: No explicit sanitization or filtering of prompt content is performed.\n- [DYNAMIC_EXECUTION]: Parallel processing is implemented using the standard library 'concurrent.futures.ThreadPoolExecutor'. No unsafe runtime code evaluation or compilation was detected.

orchestrating-agents