The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs users to create and execute shell scripts within the scripts/providers/ directory using chmod +x. It also supports discovering and running various CLI-based providers (e.g., grok-cli, kimi-cli, cursor-cli) if they are present in the system's execution path.
[EXTERNAL_DOWNLOADS]: The provided templates use curl to perform network operations against several well-known AI provider endpoints, including Google Gemini, OpenAI, Anthropic, Mistral, and Cohere. These interactions are fundamental to the skill's purpose of enabling multi-provider AI support.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a data bridge between an orchestrator and external AI services. While it employs secure JSON construction via jq --rawfile to prevent injection into the API structure, it lacks explicit sanitization or boundary markers for the natural language content it processes. This is documented as a typical surface for indirect prompt injection in agent-based integration tools.
Ingestion points: Prompts are ingested through command-line arguments or dedicated prompt files ($PROMPT_FILE) in api-patterns.md.
Boundary markers: None identified in the provided Bash templates.
Capability inventory: The scripts utilize curl for network requests, and cat or jq for local file access.
Sanitization: None applied to the natural language content; jq provides structural sanitization for JSON payloads.

provider-integration