The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill utilizes uv run python -m runtime.harness to execute a local script scripts/mcp/morph_apply.py that modifies files on the system. This provides a direct mechanism for arbitrary file writes and potential code injection into the application's codebase.
[PROMPT_INJECTION] (HIGH): This skill is highly susceptible to Indirect Prompt Injection (Category 8).
Ingestion points: The --instruction and --code_edit parameters serve as primary vectors where untrusted data enters the agent context.
Boundary markers: There are no boundary markers or sanitization protocols defined to prevent the agent from obeying instructions embedded within the code snippets it is asked to apply.
Capability inventory: The skill possesses significant file-write capabilities across the filesystem.
Sanitization: No evidence of input validation or escaping for the instructions or code edits.
[DATA_EXFILTRATION] (MEDIUM): Although not directly exfiltrating data, the ability to modify source code allows an attacker to inject 'phone-home' logic into existing scripts, which can then be used to exfiltrate environment variables or sensitive files.

morph-apply