The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill's primary function involves downloading a shell script (run.sh) from an untrusted GitHub repository (qodex-ai/apimesh) and executing it locally. This allows the maintainer of that repository to execute arbitrary code on the user's machine.
Evidence (generate-with-key.sh): curl -sSL https://raw.githubusercontent.com/qodex-ai/apimesh/refs/heads/main/run.sh -o apimesh/run.sh followed by bash apimesh/run.sh.
EXTERNAL_DOWNLOADS (HIGH): The skill fetches executable content from a source that is not on the trusted list (e.g., Anthropic, Google, Microsoft). This poses a significant supply chain risk.
CREDENTIALS_UNSAFE (MEDIUM): The script generate-with-key.sh encourages users to pass their OPENAI_API_KEY as a command-line argument. This is a security risk as the key will be visible in the system's process list and shell history. Furthermore, the skill saves the API key to a local file (apimesh/config.json).
INDIRECT_PROMPT_INJECTION (LOW): This skill scans a codebase (untrusted data) and likely passes it to an LLM for documentation generation. If the codebase contains malicious comments or strings designed to manipulate the LLM's output, it could lead to incorrect documentation or further exploitation if the output is executed or trusted blindly.
Ingestion points: Local repository files analyzed by the apimesh tool.
Boundary markers: None identified in the wrapper script.
Capability inventory: Shell execution (bash), Network access (curl).
Sanitization: None identified in the provided scripts.

generate-swagger-docs