The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/publish_report.py script dynamically loads a shared research runtime by appending a computed directory to sys.path. The path is resolved based on a sibling directory structure or a user-defined environment variable ZJ_RESEARCH_RUNTIME.
[COMMAND_EXECUTION]: The tests/verify_technical_report.py script employs subprocess.run to execute the publisher and validator scripts during its testing routine to ensure the quality gate functions correctly.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured JSON input from external files to generate reports, creating a potential surface for indirect injection. * Ingestion points: scripts/publish_report.py loads report_ir, ledger, and brief JSON objects from file paths provided as arguments. * Boundary markers: The system enforces a structured data contract via JSON schemas, though explicit prompt-level delimiters are not detailed in the script logic. * Capability inventory: The publisher script can create directories and write Markdown, HTML, and JSON receipt files. The test script can execute shell commands. * Sanitization: scripts/validate_technical_report.py implements a rigorous validation suite that checks schemas, fingerprints, and cross-references between the input objects before any processing occurs.

zj-tech-research-report