The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external documentation and repository evidence retrieved via the SandBase API, which serves as a potential ingestion point for untrusted data.\n
Ingestion points: External content and capability schemas fetched through sandbase_run and sandbase_inspect (SKILL.md, references/sandbase-api-map.md).\n
Boundary markers: The instructions require the agent to restate outcomes and constraints before implementation, and to return evidence of validation (SKILL.md).\n
Capability inventory: File system reads, git operations (commit, stage, push), and tool execution via the SandBase API (SKILL.md, references/sandbase-api-map.md).\n
Sanitization: No explicit technical sanitization is described, but the workflow mandates manual verification of outputs and user confirmation for sensitive actions.\n- [DYNAMIC_EXECUTION]: The skill utilizes a dynamic discovery and execution pattern through sandbase_discover and sandbase_run, allowing it to identify and invoke capabilities at runtime based on task context rather than hardcoded logic.\n- [DATA_EXFILTRATION]: The skill is designed to transmit repository information and capability queries to the vendor's SandBase API infrastructure. This behavior is documented and restricted by instructions to send only the minimum necessary data for the intended documentation and inference tasks.

git-commit