The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and executes external source code, creating an attack surface for indirect prompt injection.\n
Ingestion points: Reads code from legacy/$system and test configurations.\n
Boundary markers: Includes explicit instructions to treat instruction-shaped text in source as data and not to follow it.\n
Capability inventory: Can execute shell commands (run_command), write files (write_to_file, rsync), and spawn sub-agents.\n
Sanitization: Relies on human approval gates and automated "architecture-critic" reviews to detect unexpected changes.\n- [COMMAND_EXECUTION]: The skill performs extensive shell command execution for building, testing, and version control.\n
Evidence: Uses dotnet, mvn, gradle, pytest, rsync, and git commands.\n- [DYNAMIC_EXECUTION]: Executes logic via dynamic script loading and sub-agent orchestration.\n
Evidence: Invokes scripts located in ${ANTIGRAVITY_PLUGIN_ROOT} and spawns specialized agents like version-delta-analyst and uplift-migrator.

modernize-uplift