| # Skia Infra Agents Guide |
| |
| This document serves as the top-level router for agents operating within Skia |
| repositories. It contains brief descriptions of workflows and is designed to |
| keep your primary context window lean. |
| |
| **CRITICAL:** Do NOT perform complex infrastructure analysis or workflow |
| orchestration using your default behaviors. You MUST consult any relevant |
| workflows listed below before attempting these tasks. When executing these |
| workflows, you are acting as a data-gathering analyst, NOT a developer. Do NOT |
| attempt to read source code, debug, or fix issues unless explicitly instructed |
| to do so. |
| |
| ## Workflows |
| |
| Depending on your task, read details for the corresponding workflow, either by |
| running `sk agent workflow <name>` or by finding it in the `docs/` directory: |
| |
| ### task_failure_analysis |
| |
| **File:** `docs/task_failure_analysis.md` |
| **Trigger:** Use this workflow when asked to: |
| |
| - Summarize recent task failures. |
| - Investigate "what commit broke task X?" |
| - Find the root cause of recent regressions. |
| - Determine if a failing task is a hard regression or flaky infrastructure. |
| |
| ### task_drilldown |
| |
| **File:** `docs/task_drilldown.md` |
| **Trigger:** Use this workflow when asked to: |
| |
| - Provide detailed analysis of a single task. |
| - Find actual build or test failure(s), error messages, etc. |
| - Investigate flaky tasks. |
| |
| --- |
| |
| _Note: As more complex, agent-specific workflows are developed for this |
| repository, they should be added as separate markdown files in this directory |
| and linked here in this document._ |