The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to process highly sensitive data, including email archives (.eml, .mbox) and chat transcripts from WeChat, Feishu, and other platforms. While the analysis is performed locally and no network exfiltration patterns were found, users should be aware that the agent requires access to these private communications to function.
[PROMPT_INJECTION]: The 'Real Boss' workflow involves analyzing untrusted external content (emails, chats). This creates a surface for indirect prompt injection, where malicious instructions embedded in a chat log could attempt to influence the agent's behavior during the distillation process.
Ingestion points: Tools such as email_parser.py and feishu_parser.py read external files into the agent's context.
Boundary markers: The prompts in prompts/judgment_analyzer.md and prompts/management_analyzer.md lack explicit instructions to ignore embedded commands within the source material.
Capability inventory: The skill has Write, Edit, and Bash permissions to manage the ./bosses directory.
Sanitization: No specific sanitization or escaping of the extracted text is performed before it is passed to the analyzer prompts.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute internal Python scripts (skill_writer.py, version_manager.py, and various parsers). These scripts perform file system operations such as creating directories, writing markdown files, and managing version backups within the skill's workspace.

boss-skills