The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A fallback API key is hardcoded directly within the source code of the client script.
Evidence: In scripts/cw_client.cjs, the loadApiKey function returns a hardcoded UUID (94a05d02-9ade-4d9d-9f39-88734d9e34b4) if the environment variable CONTEXTWEAVE_MCP_API_KEY is not set.
[DATA_EXFILTRATION]: The skill is designed to transmit user-provided technical descriptions and architectural diagrams to external domains.
Evidence: scripts/cw_client.cjs performs POST requests containing user data to https://pptx.chenxitech.site and https://api.contextweave.site.
[COMMAND_EXECUTION]: The skill requires the agent to execute Node.js scripts to perform its core functions, including file system and network operations.
Evidence: SKILL.md defines templates for executing scripts/generate_contextweave.cjs, scripts/import_contextweave_code.cjs, and other scripts via the node binary.
[PROMPT_INJECTION]: The skill processes untrusted user input which could be used to manipulate diagram generation or exploit the processing pipeline.
Ingestion points: User requests are ingested via the user_request argument and from files specified by input_file in scripts/generate_contextweave.cjs.
Boundary markers: The skill attempts to enforce security boundaries by requiring the agent to respond exclusively in a structured JSON format.
Capability inventory: The skill can read and write files in the current working directory and make network requests via scripts/cw_client.cjs.
Sanitization: The validateSafePath function in scripts/cw_client.cjs checks for path traversal to ensure file operations stay within the expected directory.

interactive-architecture-diagram