The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_all.py uses subprocess.run to execute all Python scripts found within the .e-checker/ directory.
[REMOTE_CODE_EXECUTION]: The skill's primary workflow involves the agent generating executable Python code derived from natural language rules in checker-rule.md and then running that code. This 'generate-then-execute' behavior is a known high-risk pattern when inputs are untrusted.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection via the checker-rule.md file, which is used to influence the agent's code generation logic.
Ingestion points: The agent is instructed to find and read checker-rule.md in the work directory (Step 1 of SKILL.md).
Boundary markers: None are present; the instructions direct the agent to '逐行解析' (parse line by line) every rule without specifying delimiters or safety constraints.
Capability inventory: The skill environment allows for subprocess execution (scripts/run_all.py), file system access (Path, openpyxl), and network access (if the environment permits it, though not directly used in the templates).
Sanitization: There is no evidence of validation or sanitization of the rules before they are used to write Python scripts.

e-checker-v2