The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from several analysis files (including topology.json, ASSESSMENT.md, BUSINESS_RULES.md, and INTENT.md) to synthesize a modernization brief. This creates a surface for indirect prompt injection where malicious content in these project files could influence the agent's planning output.
Ingestion points: Files within the analysis/$system/ directory are read using view_file and incorporated into the final brief.
Boundary markers: The instructions provide a specific structure for the output brief but do not specify explicit delimiters or "ignore embedded instructions" warnings for the agent when processing the untrusted analysis data.
Capability inventory: The skill has the ability to write files (write_to_file), call other modernization tools, and execute a local shell command.
Sanitization: There is no evidence of sanitization or validation of the input file content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill executes a local Python script to finalize its output. This is a standard reporting function for the tool.
Evidence: python3 "${ANTIGRAVITY_PLUGIN_ROOT}/scripts/build_report.py" $system in SKILL.md.

modernize-brief