The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external legacy codebases provided via the $ARGUMENTS parameter. This code serves as untrusted data that could influence the behavior of subagents.
Ingestion points: The legacy codebase at $ARGUMENTS is accessed by subagents in Phase 1 (Assessment), Phase 2 (Testing), and Phase 3 (Modernization).
Boundary markers: The instructions for subagents lack explicit delimiters or "ignore instructions" directives to prevent the agents from following malicious prompts embedded within comments or documentation of the legacy code.
Capability inventory: The subagents possess extensive capabilities including code analysis, test generation, logic extraction, and infrastructure configuration (API gateways, feature flags).
Sanitization: There are no defined mechanisms to sanitize or filter the legacy code before it is processed by the AI subagents.
[DYNAMIC_EXECUTION]: The workflow relies on the dynamic generation and execution of code components, which is standard for modernization tasks but constitutes a runtime execution surface.
Evidence: The test-automator subagent generates characterization tests and test harnesses based on the legacy code's current behavior.
Evidence: The python-pro and golang-pro subagents are instructed to extract business logic and implement it using modern patterns, effectively generating new executable code units.

framework-migration-legacy-modernize