| openmetadata-endpoint | OpenMetadata API base URL (for example, https://metadata.example.com) | yes | — |
| auth-token | OpenMetadata bearer token | yes | — |
| github-token | GitHub token used to read PR diff and publish comments | no | — |
| file-patterns | Comma or newline separated glob patterns for tracked files | no | **/*.sql
**/models/**/*.yml
**/models/**/*.yaml
**/schema.yml
**/schema.yaml
**/dbt_project.yml
**/dbt_project.yaml
|
| operating-mode | Preset execution mode (fast|balanced|strict-governance) | no | balanced |
| lineage-provider | Lineage backend (api|mcp|auto) | no | auto |
| mcp-endpoint | OpenMetadata MCP JSON-RPC endpoint (defaults to {openmetadata-endpoint}/mcp) | no | — |
| max-lineage-depth | Maximum downstream traversal depth (1-3, OpenMetadata API limit) | no | 3 |
| max-concurrency | Maximum concurrent lineage lookups | no | 4 |
| max-tracked-files | Maximum tracked files analyzed per PR | no | 200 |
| max-entities | Maximum extracted entities analyzed per PR | no | 500 |
| max-downstream-assets | Maximum downstream assets traversed before truncation | no | 2000 |
| request-timeout-ms | HTTP request timeout in milliseconds | no | 15000 |
| max-retries | Maximum retries for transient API failures (0 disables retries) | no | 3 |
| max-retry-wait-ms | Maximum wait duration per retry attempt in milliseconds | no | 15000 |
| max-total-retry-wait-ms | Total retry wait budget across a single lineage request in milliseconds | no | 60000 |
| fail-on-missing-metadata | Fail action if referenced entities cannot be resolved in metadata | no | false |
| ai-summary-enabled | Enable optional AI summary layer | no | false |
| ai-summary-provider | LLM provider for AI summary (for example, openai, anthropic, google, openrouter) | no | — |
| ai-summary-model | LLM model name for AI summary (for example, gpt-4.1-mini, claude-3-5-sonnet-latest) | no | — |
| ai-summary-api-key | API key for selected AI summary LLM provider (store in GitHub Secrets) | no | — |
| strict-sql-parse | Use stricter SQL parsing mode for higher precision | no | false |
| critical-asset-tags | Comma or newline separated critical asset tags for risk scoring | no | tier1
critical
business_critical
|
| risk-high-dashboard-count | Dashboard impact count threshold for high risk | no | 5 |
| risk-high-pipeline-count | Pipeline impact count threshold for high risk | no | 4 |
| risk-high-report-count | Report impact count threshold for high risk | no | 8 |
| risk-high-total-assets | Total impacted assets threshold for high risk | no | 20 |
| risk-high-warning-count | Warning count threshold that can trigger high risk with sufficient impacted assets | no | 3 |
| risk-high-warning-min-assets | Minimum impacted assets required when warning threshold is met for high risk | no | 8 |
| risk-high-low-confidence-count | Low-confidence changed entity threshold for high risk | no | 10 |
| risk-weight-governance | Optional governance weight used in weighted risk scoring overlay | no | 0 |
| risk-weight-usage | Optional usage weight used in weighted risk scoring overlay | no | 0 |
| risk-weight-data-quality | Optional data-quality weight used in weighted risk scoring overlay | no | 0 |
| risk-weight-medium-threshold | Weighted risk score threshold that lifts low risk to medium | no | 6 |
| risk-weight-high-threshold | Weighted risk score threshold that escalates risk to high | no | 12 |
| impact-json-file | Optional path to write full structured impact JSON (for artifact upload) | no | — |
| allowed-endpoint-hosts | Optional comma or newline separated endpoint hostname allowlist | no | — |
| allow-insecure-local-endpoints | Allow http endpoints for localhost or loopback only | no | false |
| max-comment-assets | Maximum impacted assets shown per type in PR comment | no | 20 |