Agent SBTI Behavior Profile Analyzer

Problem

An agent's behavior is usually not captured by a single persona line. It is spread across SOUL.md, AGENTS.md, IDENTITY.md, and many rounds of conversation. When reviewing an agent, assigning collaboration duties, or checking behavioral consistency, engineers often lack a comparable, file-backed intermediate artifact. This skill turns role settings and recent interactions into a structured SBTI behavior profile.

How It Works

The skill first reads local workspace role files. If those are incomplete, it extracts about 20-50 representative messages from sessions_history. It then builds a five-dimension profile covering instruction response, task handling, collaboration mode, boundary awareness, and decision style, describing each dimension with a left-right polarity. After mapping the profile to SBTI answers, it calls scripts/compute_sbti_result.py to parse final.code, final.cn, final.intro, final.desc, and the poster path. The final report includes a type label, behavior summary, verbatim intro / desc, and a poster image, while a small controlled perturbation is applied to some answers to preserve the main trend.

Boundaries

The output depends on role-file completeness and session quality, so sparse context can produce a generic profile. The skill avoids leaking raw chat content, references posters only through local file paths, and requires fixed copy to be output verbatim. It is best used as a behavior snapshot or review material, not as a tool for rewriting promotional text or replacing manual security audits.