The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses natural instructional language to establish a professional PR manager persona. No evidence of malicious prompt injection, DAN-style jailbreaks, or attempts to bypass safety filters was found.
[DATA_EXFILTRATION]: No sensitive file paths, hardcoded credentials, or network exfiltration patterns were detected. The skill operates on user-provided data within the agent's text generation context.
[OBFUSCATION]: The content is provided in plain text (Thai and English) with no detected use of Base64 encoding, zero-width characters, homoglyph substitutions, or other obfuscation techniques.
[REMOTE_CODE_EXECUTION]: The skill consists solely of Markdown templates and instructions. It does not download, install, or execute external packages or scripts.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied information to generate PR drafts, which represents a potential attack surface.
Ingestion points: User arguments and the 'Story' input field in SKILL.md.
Boundary markers: None present in the templates to separate user data from instructions.
Capability inventory: Limited to text generation and markdown formatting; no tool calls or file system writes are requested in the instructions.
Sanitization: No input filtering is performed. The risk is assessed as safe because the skill lacks executable capabilities that could be exploited via injection.

pr-manager