The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides documentation patterns that are vulnerable to indirect prompt injection (specifically SQL injection) by using Python f-strings to interpolate untrusted user data into database queries. \n
Ingestion points: User input enters the context via st.text_input and st.form fields described in SKILL.md. \n
Boundary markers: No delimiters or parameterization warnings are used in the provided code examples. \n
Capability inventory: The skill uses session.sql() (via Snowpark) in both SKILL.md and connections.py to execute commands on the Snowflake database. \n
Sanitization: The examples in SKILL.md lack input validation or parameterized queries, demonstrating unsafe interpolation: session.sql(f"... WHERE date BETWEEN '{start_date}' AND '{end_date}'"). \n
[EXTERNAL_DOWNLOADS]: The skill recommends downloading packages from PyPI and Snowflake's Anaconda channel. \n
Evidence: uv pip install -r requirements.txt mentioned in SKILL.md. These target well-known package registries. \n
[COMMAND_EXECUTION]: The skill instructs the agent to execute system-level commands for package management and deployment. \n
Evidence: uv pip install --system modifies the system-wide Python environment, and snow streamlit deploy executes a deployment process via the Snowflake CLI.

streamlit-development