The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes automated assessment scripts that read and parse external codebase files, creating an attack surface where malicious content within a legacy project could attempt to influence the agent's behavior.
Ingestion points: references/system-assessment.md uses Path.rglob to discover and read all Python source files in a target directory.
Boundary markers: Absent. There are no specific delimiters or "ignore previous instructions" warnings applied to the data being read from the files.
Capability inventory: references/system-assessment.md utilizes subprocess.run (for git) and Path.write_text (for reporting).
Sanitization: Absent. The script performs static analysis via the ast module and regular expressions without sanitizing strings that might be interpreted as instructions.
[COMMAND_EXECUTION]: The assessment logic in references/system-assessment.md performs legitimate shell command execution to analyze project history.
Evidence: subprocess.run(['git', 'log', '--format=format:', '--name-only'], ...) is used to determine which files are most frequently changed to identify maintenance hotspots.
[DATA_EXFILTRATION]: The migration templates include code for establishing network connections, which is a potential surface for data movement.
Evidence: references/migration-strategies.md provides a proxy implementation using httpx.AsyncClient to route traffic between legacy and modern service endpoints.

legacy-modernizer