The Agent Skills Directory

[SAFE]: The skill operates entirely within local project directories and focuses on summarizing experimental data. It does not perform network operations or access sensitive system configuration files like SSH keys or cloud credentials.
[SAFE]: The skill includes a robust security design that prevents autonomous execution of high-impact actions. It uses a human-in-the-loop mechanism (Gate G4.5) ensuring that method selection, result verdicts, and the archival of source code only occur after a human user reviews and updates a decision artifact to a 'DECIDED' status.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes untrusted data from experimental outputs, including CSV files, JSON data, metrics, and logs in the results/Qx/experiments/roundN/ directory.
Ingestion points: Raw experimental output files, logs, and run_summary.json files are read into the context in Workflow Steps 1 and 2.
Boundary markers: The skill uses specific text tags like [AI-SUGGESTED — not a verdict] and [PENDING-MODELER] to distinguish AI interpretations from authoritative data, which helps prevent the agent from treating injected instructions in data as system commands.
Capability inventory: The skill has the ability to read and write files, and to move code files from the code/ directory to workspace/archived/ based on decisions.
Sanitization: There is no explicit mention of sanitizing or escaping the content of processed data files, but the strong human gate mitigates the risk of automated exploitation.

result-report-generator