The Agent Skills Directory

[SAFE]: The skill is a set of instructional documents and templates for Snowflake MLOps. It does not contain executable malicious code.
[REMOTE_CODE_EXECUTION]: While the skill provides a GitHub Actions template (github-actions-deploy.yml) that includes shell commands and package installations (pip install), these are intended for the user to implement in their own CI/CD environment. The packages listed (snowflake-connector-python, snowflake-snowpark-python, snowflake-ml-python) are official and trusted Snowflake libraries. There is no evidence of the agent executing untrusted remote code on the host.
[CREDENTIALS_UNSAFE]: The skill explicitly promotes secure credential management, recommending the use of OIDC/Workload Identity Federation and service accounts to avoid storing secrets in CI/CD pipelines.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface in Category 8 as it ingests user input to determine MLOps maturity and generate implementation checklists. However, it uses these inputs to route the user through predefined, safe instructional content in markdown files. There is no unsafe interpolation of untrusted data into executable commands within the agent's own context.
Ingestion points: User intent and project context parameters (maturity level, environment names) ingested in SKILL.md and INSTRUCTIONS.md.
Boundary markers: The instructions use structured routing logic and explicit sub-flows to maintain control over the session state.
Capability inventory: Uses Bash, snowflake_sql_execute, Read, Write, Edit, Glob, and Grep tools.
Sanitization: The skill relies on static reference files for the bulk of its output, minimizing the risk from dynamic content generation.

mlops