The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's analysis scripts process external data files, creating a potential surface for indirect prompt injection if the simulation results contain malicious instructions.
Ingestion points: scripts/agent.py reads user-provided CSV files via the --simulation-csv and --training-csv flags. scripts/process.py reads user-provided JSON files via the --data argument.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are present in the scripts or the SKILL.md instructions.
Capability inventory: The scripts perform file reads, data manipulation using pandas and numpy, and local file writes for reporting.
Sanitization: The scripts use standard pandas.read_csv and json.load methods. While these prevent direct code execution, they do not sanitize textual data that could influence an LLM's behavior if the analysis results are returned to the agent's context.

implementing-anti-phishing-training-program