The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (search_user.py and work_list.py) to interact with external APIs. These scripts explicitly disable SSL certificate verification using ssl.CERT_NONE, which exposes the communication to man-in-the-middle (MITM) attacks.
Evidence (scripts/search_user.py): ctx.verify_mode = ssl.CERT_NONE (line 64).
Evidence (scripts/work_list.py): ctx.verify_mode = ssl.CERT_NONE (line 68).
[PERSISTENCE]: The skill uses the automation_update tool to create recurring daily tasks that execute shell commands (python3). While intended for the "subscription" feature, this provides a mechanism for persistence and automated command execution.
Evidence (SKILL.md): Step 6 describes creating tasks with prompt templates that run the work_list.py script daily at 9:00.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the Dongchedi platform (user bios, work titles, descriptions) and interpolates it into the agent's context.
Ingestion points: API responses from redfox.hk parsed in search_user.py and work_list.py.
Boundary markers: The skill relies on standard Markdown table formatting but lacks explicit security delimiters for external data.
Capability inventory: Execution of Python scripts via shell and task scheduling via automation_update.
Sanitization: The scripts include an _escape_markdown function to prevent external content from breaking the agent's table rendering or link structures.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to the vendor's domain (https://redfox.hk) to retrieve user and work data.

dongchedi-user-search