debuglisted

# swarm:debug — from intent, to a confirmed root cause, to a dispatched fix > Read `swarm:orchestrate` first for fan-out mechanics and the blinded-verification rule. This is the **debug loop**: pin what you *meant* against what *happened*, find the gap in the contract that let it slip, prove the root cause with independent agents, then close it out — artifact, ticket, dispatch. No fix ships until the cause is proven and independently confirmed. You are debugging: **$ARGUMENTS** The root cause is **not** where the error surfaces — it's where the incorrect behavior originates. No fallbacks, no "just in case" guards, no symptom patching — fix the root cause or fix nothing. ## 1. Intent vs observed — pin the delta before touching code Separate **intent** from **observation** first; a root cause that doesn't explain the delta is the wrong root cause. If the report came with screenshots, logs, or a recording, read them (a `host:/path` clip is a real file — resolve and read it, don't treat it as literal text). State, one line each: - **Intended** — what the user was trying to do. - **Observed** — what actually happened: the exact surface, message, or state. - **Delta** — the specific gap between them. This frames everything below. If intent is genuinely ambiguous from the report plus the code, ask once. Otherwise infer it and proceed. ## 2. Check the spec — find the gap that let it slip Before diagnosing, find the capability's contract and its tests. "Why did it slip" is a