fix-flowlisted

# fix-flow — Safe project fix workflow Codifies the global CLAUDE.md discipline (`tools first, grep last` · impact before change · verify · record correction) into a repeatable 5-step fix loop. This is the **execution** guardrail — it assumes you already know WHAT to fix. If the root cause is unknown, run `systematic-debugging` first, then come back here to apply the fix. ## When to use vs not | Situation | Use | |-----------|-----| | Known fix to apply in a project | **fix-flow** (this) | | Unknown root cause / mysterious bug | `systematic-debugging` first | | Same edit across 5+ files (regex) | `bulk-fixer` | | Enum/status drift across layers | `enum-sync-checker` | | Pre-deploy gate | `pre-deploy-check` | ## The 5 steps (do them in order, don't skip) ### 1. UNDERSTAND — pin the target - Restate the bug in one line + the exact symptom (error text, wrong behavior). - Define "done": the concrete, verifiable success condition. - Read the project graph FIRST (cheap context): `{{WIKI_PATH}}\<Project>\graph\knowledge_graph.json` — check `critical_rules` for a known gotcha before touching code. ### 2. IMPACT — what will break (BEFORE editing) - GitNexus MCP: `impact({target: "<symbol/function>", minConf: 0.8})`. 1 `impact()` ≈ 500 tokens, replaces ~10 Grep+Read (~50k tokens). Always prefer. - If GitNexus has no data for the repo, fall back to a single targeted Grep for callers. - Output: list the call sites / dependents the fix could affect. If the blast radius is larg