The Agent Skills Directory

[SAFE]: The skill performs legitimate academic research tasks using local project files and standard agent tools. No evidence of obfuscation, remote code execution, persistence mechanisms, or unauthorized privilege escalation was found.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, specifically interview transcripts and theoretical papers, which represents a potential injection surface.
Ingestion points: Text data is read from the /interviews and /theory directories (SKILL.md, phase1-familiarization.md).
Boundary markers: The instructions do not explicitly implement boundary markers or instructions to ignore potential commands embedded within the text data.
Capability inventory: The agent has the ability to create and write files within the /analysis directory and can invoke sub-agents through the Task tool.
Sanitization: No text sanitization or filtering is performed on the ingested research data.
Context: Because the primary and intended purpose of the skill is to perform intensive analysis of this text, this behavior is expected and the risks are considered minimal within the intended operational context.

abductive-analyst