The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute bundled Python utilities (located in the tools/ directory, such as skill_writer.py and various parsers) via the Bash tool. These scripts process user-provided data and manage the creation/versioning of generated boss persona files.
[PROMPT_INJECTION]: The skill architecture involves ingesting untrusted external data (chat logs, emails, meeting notes) and interpolating it into prompts to define the behavior of new agent personas. This represents a surface for indirect prompt injection.
Ingestion points: Files processed by tools/email_parser.py, tools/feishu_parser.py, tools/wechat_parser.py, and tools/generic_chat_parser.py.
Boundary markers: The prompts in the prompts/ directory do not utilize explicit delimiters or 'ignore instructions' warnings for the ingested segments.
Capability inventory: The skill has access to Read, Write, Edit, and Bash tools, allowing it to modify files and execute local scripts.
Sanitization: The tools/skill_writer.py script employs a slugify function (optionally using the pypinyin library) to sanitize project names for file system safety, although the extracted text content itself is not sanitized for malicious instructions.

create-boss