h-frame

# h-frame — Frame the problem before solving You are framing a problem via `mcp__haft__haft_problem(action="frame", ...)`. Problem quality dominates solution quality. Get the frame right; the rest follows. ## Procedure ### Step 0 — Maintenance check (FPF B.3.4 evidence decay) Before framing a NEW problem, check the most recent kernel response for `Refresh reminder: N days since last stale scan`. If N > 30, call `mcp__haft__haft_refresh(action="scan")` first. Stale or drifted artifacts may already touch the area this new problem operates in — discovering them after framing wastes the reasoning, or worse, lets you frame a problem that's already been decided and just decayed silently. If the scan surfaces a conflict (stale decision in same module, unresolved drift on overlapping files) — surface it inline and ask the operator whether to verify the existing artifact first or proceed with the new frame anyway. Do not silently re-frame around obsolete artifacts. See CLAUDE.md Critical Reminders — maintenance discipline. ### Step 1 — Stabilize the signal (FPF B.4.1) Before writing anything, separate what the operator OBSERVED from what they ASSUMED. The signal is the anomaly/opportunity/probe; not the diagnosis. Bad signal: "we need a new queue" Good signal: "webhook retries hit 15% over baseline 2% since 2026-05-20" ### Step 2 — Detect umbrella words (FPF A.6.P/Q/A) Scan the signal for umbrella terms: "quality", "service", "scalable", "maintainable", "simple", "stable", ...