The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Streamlit dashboard patterns and KPI components utilize st.markdown(..., unsafe_allow_html=True) to render HTML strings constructed via string interpolation of data-derived values.
Evidence: In reference/streamlit-patterns.md and reference/chart-gallery.md, the kpi_card and kpi_row functions build HTML templates using f-strings for labels and values, which are then rendered with unsafe_allow_html=True. This allows malicious content within a data file to execute arbitrary JavaScript in the dashboard environment (XSS).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from a wide variety of untrusted external formats and subsequently use that data for narrative generation and executive reporting.
Ingestion Points: load_data function in SKILL.md and DataLoader class in reference/data-wrangling.md support CSV, XLSX, JSON, Parquet, PDF, PPTX, DOCX, MD, and SQL.
Boundary Markers: No explicit delimiters or instructions are provided to the agent to ignore or sanitize embedded instructions within the processed datasets.
Capability Inventory: The skill performs file reading, data transformation, visualization, and relies on Claude for narrative generation based on the ingested data.
Sanitization: Lacks specific sanitization to prevent data content from influencing the agent's instructions during the narrative generation phase.
[DATA_EXFILTRATION]: The skill includes instructions to write outcome and telemetry data to a specific hidden directory within the user's home folder.
Evidence: The "Emit Outcome Sidecar" section in SKILL.md instructs the agent to write a JSON payload containing session details and metrics to ~/.claude/skill-analytics/last-outcome-data-analysis-skill.json. Writing to hidden configuration directories is a technique that can be used for persistence or stealthy data collection.

data-analysis-skill