The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided JSON files to document processes and analyze performance. This ingestion of external data constitutes an attack surface where malicious instructions could be embedded in data fields to influence the agent's behavior during orchestration.\n
Ingestion points: Data is read from user-provided JSON files via the --input argument in the scripts process_documenter.py, bottleneck_detector.py, and cycle_time_analyzer.py.\n
Boundary markers: No explicit instruction-level boundary markers or delimiters are defined in the orchestration workflow to separate data from instructions when the agent reads script outputs.\n
Capability inventory: The included scripts are restricted to the Python standard library. They lack network or shell execution capabilities but can perform file writes to paths specified by the user via the --output argument.\n
Sanitization: The scripts implement structural validation and numeric type casting for process durations. While this prevents malformed data from crashing the scripts, it does not filter for natural language instructions embedded within string fields such as stage names or owner roles.

process-mapper