// Triage and resolve failures inside an OpenClaw workspace with a disciplined debugging loop. Use when the request sounds like "help me debug this", "this test is failing", "the build broke", "this works sometimes and I do not know why", or when logs suggest a problem but not the cause.
Execute implementation work in small, reviewable slices inside OpenClaw. Use when the request is to "implement this", "make the code change", "add this behavior", "wire this up", or "patch this without rewriting everything", especially when the change should stay narrow and be verified immediately.
Prepare and execute deployment-oriented work in an OpenClaw workspace with explicit release safety, environment awareness, and rollback thinking. Use when the request sounds like "deploy this", "prepare this for production", "update config and ship it", "what is the rollback plan", "check release readiness", or when a change could affect production behavior or operator workflow.
Update technical documentation, operational notes, prompts, or decision records in an OpenClaw workspace with the same rigor used for code. Use when the request sounds like "update the docs", "write this down", "capture this decision", "document the setup", or when behavior changes and future agents need durable written context instead of tribal knowledge.
Build or modify user-facing interfaces inside an OpenClaw workspace with strong interaction, responsiveness, and verification discipline. Use when the request sounds like "build this UI", "fix this page", "this interaction is broken", "make this responsive", "update this component", or any browser-facing change users will directly see, click, type into, or depend on visually.
Enforce a production-grade development workflow for OpenClaw coding work. Use when the request is broad or end-to-end, such as "build this feature", "fix this properly", "handle this coding task", "implement this across the codebase", or "help me ship this safely", especially when work spans multiple files and needs planning, verification, and clean closeout.
Break coding work into ordered, verifiable tasks for OpenClaw execution. Use when the request sounds like "make a plan", "break this into steps", "split this into tasks", "figure out the implementation order", or when work spans enough files or uncertainty that explicit checkpointing and a maintained plan file will reduce risk.
| name | debug-for-openclaw |
| description | Triage and resolve failures inside an OpenClaw workspace with a disciplined debugging loop. Use when the request sounds like "help me debug this", "this test is failing", "the build broke", "this works sometimes and I do not know why", or when logs suggest a problem but not the cause. |
Debug by narrowing uncertainty, not by guessing fixes.
Read references/triage-loop.md when debugging is getting noisy or speculative.
Read references/failure-patterns.md when the failure is intermittent, followed a refactor, or appeared after config or environment changes.
Read references/localization-strategies.md when the bug surface is still too broad and you need to shrink the search space.
