The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/qunfa_lint.py performs dynamic path manipulation by modifying the sys.path variable at runtime. It computes import paths relative to its own file location (e.g., parents[5] / "src" and parents[3] / "tools") to load the internal siyu_team library. This is a common pattern in complex local repositories but falls under dynamic loading from computed paths.
[COMMAND_EXECUTION]: SKILL.md contains instructions for the agent to execute shell commands using python3 to run internal scripts like qunfa_lint.py and atoms_query.py. These scripts process text generated by the agent or provided by the user, which requires the execution environment to handle shell escaping correctly.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user inputs and the assets/真实语料-待填.md file to construct group messaging copy, creating a surface for potential indirect prompt injection.
Ingestion points: Promotional details (brand, category, discounts) provided by the user and the contents of assets/真实语料-待填.md.
Boundary markers: The skill instructions lack explicit delimiters or instructions to ignore embedded commands within the interpolated data.
Capability inventory: The skill can execute local scripts and read files from the local filesystem.
Sanitization: A compliance scanner (qunfa_lint.py) is used to check for marketing policy violations, but it is not a dedicated security sanitizer for filtering malicious injection payloads.

siyu-qunfa