Agent skill · software engineering · thienanblog
debugging-workflow
Coordinator-routed specialist for reproducing and isolating unexplained failures before fixing the confirmed cause. Use after project-development-mindset routes root-cause work here, or directly when explicitly invoked or installed standalone. Do not use for known-cause fixes, routine implementation test failures, test strategy, or an established performance bottleneck.
Why this skill is useful
Provides a structured debugging workflow that includes specific steps and documentation practices not commonly found in generic resources.
What it needs
About 2k tokens when loaded. Last updated 2026-08-05. 63 stars on the source repository.
What this skill does
Debugging Workflow Use this skill when the task starts from a failure or unexplained behavior. Preserve business intent, isolate the smallest failing case, and fix the confirmed cause. Run this skill in the main conversation. Do not spawn subagents, agent teams, or delegated parallel workers unless the user explicitly approves the proposed count and scope after being told that doing so can increase usage. Ask again before expanding an approved scope. Operating Rules Reproduce before fixing when possible. Read logs, stack traces, test output, screenshots, and source-of-truth docs before changing code. Do not replace unusual business logic with a generic solution. Do not make broad refactors while the failure cause is still unknown. Form one hypothesis at a time and test it with the fastest reliable check. Keep temporary debug code local and remove it before completion. Add or update a regression test when practical. If the root cause is performance-related, return routing control to project-development-mindset and replace this workflow with performance-optimization when available. If the main work becomes test strategy, return routing control to the coordinator and replace this workflow with testing-verification when available. Workflow 1. Capture The Failure Record the exact command, URL, input, account state, branch, environment, and error output. Preserve stack traces and failing assertions. For UI bugs, capture the target element/region screenshot first, then full page only if needed. For user screenshots that are ambiguous, annotate unclear regions with labels and ask before coding. 2. Read Source Of Truth Read AGENTS.md, CLAUDE.md, relevant docs, feature specs, design-system docs, route/API/schema files, and nearby tests. Identify whether current behavior is wrong or the expectation is unclear. If docs and code disagree, call out the conflict instead of silently choosing one. 3. Reproduce And Minimize Run the smallest command or flow that exposes the failure. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills thienanblog/debugging-workflow