The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code from various directories using Large Language Models (LLMs). This functionality introduces a surface for indirect prompt injection attacks, where malicious content within the files being scanned could attempt to override the AI's instructions or bias its vulnerability reports. • Ingestion points: Processes files in user-specified directories (e.g., src/), package.json, and dependency trees. • Boundary markers: No explicit mention of delimiters or instructions to ignore embedded commands within the scanned content. • Capability inventory: Reads local files, executes external commands like npm audit, and interacts with the Anthropic API. • Sanitization: No evidence of sanitization or filtering of the scanned content before it is passed to the AI model.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install the tool and its dependencies via external registries such as npm and clawhub. While npm is a well-known service, the use of clawhub as a third-party installation source warrants user caution regarding package integrity.

flaw0