The Agent Skills Directory

[SAFE]: The skill provides structured guidance for IRB (Institutional Review Board) compliance in Taiwan. All instructions and templates are benign and intended for educational and academic research purposes.
[PROMPT_INJECTION]: No evidence of prompt injection or bypass instructions detected. The logic focuses entirely on research ethics evaluation.
[DATA_EXFILTRATION]: No network operations (curl, wget, etc.) or access to sensitive local file paths (e.g., .ssh, .env) were found. The skill does not attempt to send data externally.
[REMOTE_CODE_EXECUTION]: No remote script downloads or dynamic code execution patterns (eval, exec) are present.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests user-provided research descriptions to generate reports. However, it lack dangerous capabilities like automated network exfiltration or command execution based on that input.
Ingestion points: Step 1 in SKILL.md asks for user research designs and data collection methods.
Boundary markers: Absent.
Capability inventory: The skill allows Bash, Read, Write tools in its configuration, but no scripts in the provided files utilize them for dangerous operations.
Sanitization: No explicit sanitization or filtering instructions for user input are defined.

tw-research-ethics-reviewer