The Agent Skills Directory

[COMMAND_EXECUTION]: The framework includes a CodeAct agent designed to synthesize and execute Ruby code dynamically at runtime.
Evidence: Documentation in references/core-concepts.md describes CodeAct as an agent that 'synthesizes and executes Ruby code'.
[COMMAND_EXECUTION]: The skill provides toolsets that allow the agent to execute system commands and database operations.
Evidence: references/toolsets.md describes GitHubCLIToolset for GitHub CLI interactions and DatabaseToolset for executing SQL commands.
[PROMPT_INJECTION]: The framework establishes a surface for indirect prompt injection by processing external data with high-capability agents.
Ingestion points: Data inputs in DSPy::Signature classes as seen in SKILL.md and references/core-concepts.md (e.g., email_content).
Boundary markers: Relies on structured output formatting and JSON Schema to delimit data.
Capability inventory: Dynamic code execution (CodeAct) and shell command execution (GitHubCLIToolset).
Sanitization: The provided templates do not show explicit validation or sanitization of input data before processing.
[EXTERNAL_DOWNLOADS]: The documentation references several official adapter gems for connecting to LLM providers.
Evidence: SKILL.md and references/providers.md mention gems such as dspy-openai, dspy-anthropic, and dspy-gemini.
[DATA_EXFILTRATION]: The framework supports exporting runtime telemetry to the Langfuse service.
Evidence: references/observability.md describes the observability configuration used to send traces and metrics to Langfuse.

dspy-ruby