The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It autonomously discovers and ingests untrusted data from external websites (API documentation) via tools like WebSearch and WebFetch. This untrusted content is used to define API structures and base URLs which directly influence high-capability actions such as writing files and executing shell commands.\n
Ingestion points: Step 2: Documentation Discovery and Step 9: Deep Research involve fetching external documentation URLs.\n
Boundary markers: No instructions are provided to use delimiters or ignore embedded instructions within the researched content.\n
Capability inventory: The skill uses Bash (for curl requests), Write, and Edit (to create the connectivity-schema folder and OpenAPI specifications).\n
Sanitization: The instructions do not direct the agent to sanitize or validate strings sourced from external documentation before they are interpolated into shell command strings.\n- [COMMAND_EXECUTION]: In Step 13.2.1, the skill constructs shell commands for the Bash tool using data retrieved from external research. Parameters such as the Base URL and endpoint Path are interpolated directly into curl command lines. If the documentation being researched contains malicious shell metacharacters, it could result in arbitrary command execution when the validation loop runs.\n- [SAFE]: The skill implements a robust 'Credential Security' framework that prevents sensitive authentication values from being exposed to the AI model's context. It uses shell-level substitution and grep -c for existence checks, ensuring that secrets remain localized to the execution environment.

generate-connectivity-knowledge