The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to research and evaluate external GitHub repositories and sources, creating a surface for indirect prompt injection where untrusted external content could attempt to manipulate agent behavior.
Ingestion points: External data retrieved through search, scraping, and enrichment activities as described in SKILL.md and executed via sandbase_run.
Boundary markers: While the skill emphasizes recording provenance and cross-checking claims, it lacks instructions for using explicit delimiters (e.g., XML tags or triple backticks) to separate untrusted external content from the system prompt.
Capability inventory: The skill utilizes sandbase_run for executing capabilities, as well as sandbase_discover and sandbase_inspect for capability metadata management.
Sanitization: No specific technical sanitization, filtering, or escaping mechanisms for external content are detailed in the instructions.
[COMMAND_EXECUTION]: The skill provides instructions for utilizing a suite of sandbase_* tools to execute external tasks. These tools are associated with the vendor's own infrastructure and are governed by strict instructions requiring user confirmation for costs and the use of schema-validated arguments.

github-gem-seeker