The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs command execution by calling python3 fdk/tools/medic.py to run various system health probes. This action is the primary function of the skill but introduces a dependency on the integrity of the local script and the security of the parameter handling.
[INDIRECT_PROMPT_INJECTION]: The skill relies on natural language input from the user to define the scope of the diagnostic check, which is then interpolated into a shell command: python3 fdk/tools/medic.py [phạm vi]. This lack of strict validation or sanitization instructions for the user-supplied string creates a command injection surface.
Ingestion points: User input defining the scope of probes (e.g., 'rules', 'docs', 'eval') as described in the 'Input và output contract' section.
Boundary markers: There are no defined delimiters or specific instructions for the agent to encapsulate or escape the interpolated string.
Capability inventory: The skill utilizes shell command execution via the python3 binary.
Sanitization: No sanitization, allow-listing, or validation logic is specified in the workflow to prevent malicious input from being passed to the shell.
[METADATA_POISONING]: The skill's documentation contains internal inconsistencies regarding its impact on the system. While the 'Capabilities' section claims the tool is read-only ('không ghi'), 'RULE-05' and 'Branch B01' mention 'regen/sync' actions and providing 'fix' commands for the user to execute. This inconsistency could lead a user or agent to misjudge the skill's potential for modifying the repository.

medic