The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The framework supports dynamic code synthesis and execution through the dspy-code_act gem.
Evidence: Documentation in references/core-concepts.md describes a "Think-Code-Observe" agent that executes synthesized Ruby code.
[COMMAND_EXECUTION]: The skill includes a toolset for interacting with the GitHub CLI (gh).
Evidence: references/toolsets.md defines the GitHubCLIToolset, which executes shell commands to perform GitHub operations like listing issues or making API requests.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, creating a surface for indirect prompt injection.
Ingestion points: The EmailClassifier signature in SKILL.md and various text-processing tools in references/toolsets.md ingest raw strings from external sources.
Boundary markers: The framework uses JSON schemas and structured outputs, but lacks explicit instructional delimiters to isolate untrusted content from system instructions.
Capability inventory: The agent possesses high-impact capabilities including dynamic Ruby execution (CodeAct) and GitHub CLI access.
Sanitization: While the framework enforces Sorbet types and filters internal metadata discriminators, it does not provide built-in sanitization for natural language content ingested into prompts.

dspy-ruby