← ClaudeAtlas

diagnoselisted

Disciplined diagnosis loop for hard bugs and performance regressions. Build a feedback loop → reproduce → minimise → hypothesise → instrument → fix → regression-test. Use when user says "diagnose this" / "debug this", reports a bug, says something is broken/throwing/failing, describes a performance regression, or hands you an issue summary from a previous session.
SrdjanCoric/skills · ★ 18 · AI & Automation · score 64
Install: claude install-skill SrdjanCoric/skills
# Diagnose A discipline for hard bugs. Skip phases only when explicitly justified. ## Input Can start from a fresh session. Look for a **bug handoff** from a previous session — written by the `handoff` skill at `handoffs/<branch>/<slug>.md` (or a path the user passes). If one exists, read it first and map its sections straight onto this skill: - **What we've tried & ruled out** → pre-seeds your discarded hypotheses. Do **not** re-test these. - **Current hypothesis** → seeds Phase 4. - **Relevant files & locations** → where to start looking. A good handoff lets you skip rediscovery and jump toward the feedback loop and hypotheses. With no handoff, treat whatever the user gives you as the symptom report — what was tried, what failed, the exact symptom. When exploring the codebase, use the project's domain glossary to get a clear mental model of the relevant modules, and check ADRs in the area you're touching. ## Phase 1 — Build a feedback loop **This is the skill.** Everything else is mechanical. If you have a fast, deterministic, agent-runnable pass/fail signal for the bug, you will find the cause — bisection, hypothesis-testing, and instrumentation all just consume that signal. If you don't have one, no amount of staring at code will save you. Spend disproportionate effort here. **Be aggressive. Be creative. Refuse to give up.** ### Ways to construct one — try them in roughly this order 1. **Failing test** at whatever seam reaches the bug — unit, integration, e2e.