The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strong instructional constraints designed to enforce persona fidelity (e.g., 'REQUIRED READ', 'MUST read persona.md'). These are standard pedagogical techniques for expert simulation and do not attempt to bypass safety filters, extract system prompts, or override agent safety protocols.
[DATA_EXFILTRATION]: No sensitive file paths (e.g., .ssh, .aws), network operations (e.g., curl, wget), or hardcoded credentials were detected. The skill operates exclusively on the provided local data.
[OBFUSCATION]: No Base64, hex-encoding, zero-width characters, or homoglyph attacks were identified. The text is plain, human-readable markdown across all files.
[REMOTE_CODE_EXECUTION]: The skill does not utilize package managers (npm, pip) or fetch external scripts. It relies solely on internal file references to ./persona.md and the ./topics/ directory.
[DYNAMIC_CONTEXT_INJECTION]: No use of the dynamic exclamation-backtick (!command) syntax was found in the SKILL.md or supporting files.
[INDIRECT_PROMPT_INJECTION]: While the skill processes user-supplied questions to generate expert advice, it defines clear output boundaries for separate experts and focuses on internal framework application. No exploitable capability-data chains were identified.

dojo-peter-thiel