The Agent Skills Directory

Command Execution (MEDIUM): The 'Verification Scripts' section contains instructions for the agent to execute a variety of Python scripts located in the .agent/skills/ directory (e.g., python .agent/skills/vulnerability-scanner/scripts/security_scan.py .). This pattern involves executing local code that is not contained within the skill itself, relying on the security and integrity of the broader agent environment.
Indirect Prompt Injection (LOW): The 'Script Output Handling' instructions mandate that the agent 'capture ALL output', 'parse the output', and 'summarize to user'. This creates a vulnerability surface where a validation script—which may be processing code or data influenced by an attacker—could produce output containing malicious instructions that the agent might inadvertently follow during its summarization and decision-making phase.
Ingestion points: Script output captured by the agent after running verification commands.
Boundary markers: None specified for the script output; the agent is simply told to capture and parse 'ALL output'.
Capability inventory: The skill allows 'Read', 'Write', and 'Edit' tools, and explicitly instructs the execution of shell commands via Python.
Sanitization: No evidence of sanitization or filtering of the script output before the agent processes and summarizes it.

clean-code