The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted legacy source code and external documentation to guide its modernization logic.
Ingestion points: The agent reads legacy code from legacy/, as well as analysis/$system/MODERNIZATION_BRIEF.md, BUSINESS_RULES.md, and RULE_REVIEWS.json.
Boundary markers: The instructions tell the agent that the MODERNIZATION_BRIEF.md is "binding," but there are no explicit delimiters or sanitization steps defined for these external inputs.
Capability inventory: The skill allows the agent to execute shell commands, run compilers and test frameworks (like Maven, NPM, or Pytest), write files, and spawn subagents.
Sanitization: While the skill contains strong instructions to mask credentials and avoid production targets, it does not explicitly sanitize the logic extracted from the input files.
[COMMAND_EXECUTION]: The skill invokes various system tools and internal scripts to perform its task.
Evidence: It executes environment checks (java -version, node -v, python3 -V) and build tools (mvn, npm, pytest).
Internal Scripts: It executes scripts from a plugin root variable: ${ANTIGRAVITY_PLUGIN_ROOT}/scripts/compare.py and ${ANTIGRAVITY_PLUGIN_ROOT}/scripts/build_report.py.
[DYNAMIC_EXECUTION]: The skill's core workflow involves generating new source code and unit tests at runtime and subsequently executing them.
Evidence: The agent is instructed to write implementation and test files to the modernized/ directory and then run the test suite to prove behavioral equivalence (Step 3).

modernize-transform