The Agent Skills Directory

[SAFE]: The skill instructions and agent configuration do not contain any evidence of malicious intent, unauthorized command execution, or data exfiltration. The described tool usage for memory persistence is governed by clear technical constraints and user-endorsement requirements.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it is designed to analyze external data such as code and logs. 1. Ingestion points: The Literal Read procedure in SKILL.md extracts components and flows from code, tests, and repository structures. 2. Boundary markers: No specific delimiters are defined in the instructions to isolate this ingested content. 3. Capability inventory: The skill can write data to a memory extension using the run_memory_note_tool. 4. Sanitization: No explicit sanitization or escaping of external content is specified, however, the skill mandates that metaphors never substitute for literal proof such as tests or logs.

synesthesia