The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled data from files such as inference.csv, experiment.yaml, and status.json across all five investigation phases. Maliciously crafted data in these files could attempt to influence the agent's reasoning or report generation. This is an inherent risk for analysis tools ingesting external datasets.
[DYNAMIC_EXECUTION]: The skill generates and executes Python code through the Bash tool (e.g., in hooks/rca-script-check.sh and Phase 2B/3A instructions) to calculate image intensities, perform threshold sweeps, and compute component-level statistics. These scripts are dynamically created to handle the specific structure of the provided experiment results.
[DATA_EXPOSURE]: The hooks/rca-package.sh script is designed to access platform-specific metadata and configuration files, such as ~/.claude/settings.json and session transcripts, to bundle them into the rca_results directory. While this facilitates reproducibility, it involves the movement of configuration data within the local workspace.

tao-analyze-changenet-rca