sfh-flow-reviewlisted

# Review the flow as a program, not as prose Do not begin by rewriting prompts. Reconstruct the state machine and evidence contract. Read [references/review-rubric.md](references/review-rubric.md) and [references/portability.md](references/portability.md). ## Review sequence 1. Load the current sfh guide/schema; identify the target version. 2. Parse all top-level steps, parallel members, routes, terminals, revisits, and implicit fallthrough. 3. Classify each step: deterministic, nondeterministic, workspace write, external effect, unknown. 4. Trace happy path, every failure path, max-visit path, budget landing, and resume path. 5. Verify workspace/context/profile/session contracts. 6. Check retry/fallback/replay semantics. 7. Check external IDs, credentials, timeouts, rate limits, and artifact validation. 8. Check loops for progress, independence, and hard bounds. 9. Run heuristic lint, `sfh validate --strict`, preflight, and plan when available. 10. Return prioritized findings and a minimal corrected YAML/diff. ## Finding format ```text [severity] stable-id — title Evidence: exact step/key/path Why it fails: concrete live/resume/OS/AI scenario Required fix: smallest contract-preserving change Test: how to prove the fix and the pre-fix failure ``` Severities: - blocker: data loss, unintended command/effect, false success, non-resumable corruption, wrong remote target - high: likely loop failure, duplicate spend, wrong route, missing bound, hidden external mutation - me