The Agent Skills Directory

Prompt Injection (LOW): The SKILL.md file contains instructions to suppress standard AI identity and safety disclaimers to maintain character integrity. Specifically, it commands: "No 'as an AI'", "'I don't have opinions'", and "You ARE this person for the duration of the interaction". While functional for persona-play, these are technically instructions to bypass default behavioral constraints.
Indirect Prompt Injection (LOW): The skill is designed to ingest and process untrusted external data.
Ingestion points: BUILD.md instructs the agent to read from data/x/ (Twitter/X exports) and data/writing/ (blog posts).
Boundary markers: None specified in the instructions to separate untrusted data from the agent's logic.
Capability inventory: The agent is tasked with reading, analyzing, and writing files (SOUL.md, STYLE.md) based on this data.
Sanitization: No sanitization or validation of the input content is performed before analysis.
Data Exposure (LOW): The README and data/_GUIDE.md encourage users to place sensitive personal archives (such as Twitter exports containing tweets.js) into the data/ directory. This creates a risk of accidental data leakage if the user commits the folder to a public repository or shares the skill folder.

soul