The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data sources such as CSV files, Excel spreadsheets, and SQL tables. This constitutes an attack surface where malicious instructions could be embedded in the data. However, the skill significantly mitigates this risk by providing extensive instructions for data quality validation, schema inspection, and data governance.
Ingestion points: Processes datasets provided by users or retrieved from databases in Phase 2 of the workflow.
Boundary markers: Utilizes a 'Cognitive Separation' framework to distinguish raw data (Observations) from validated Facts and Hypotheses.
Capability inventory: Performs SQL aggregation, programmatic data wrangling (e.g., pandas/polars), and statistical inference. No arbitrary shell command execution or network exfiltration capabilities are implemented.
Sanitization: Explicitly mandates PII protection, hashing, masking, and data minimization techniques in the reproducibility-and-governance.md reference.
[SAFE]: The skill instructions and reference materials focus entirely on standard analytical best practices. There is no evidence of prompt injection, obfuscation, hardcoded credentials, or unauthorized remote code execution. The workflow promotes transparency, auditability, and human-in-the-loop verification.

data-analyst