Review

Binding sources

• docs/FINAL_SYSTEM_VISION.md — 10-layer vision (L5 AI support; L6 learning/optimization).
• AGENTS.md — enforceable boundaries; verify bounded autonomy, no black-box paths, strategy quality signals, L7 ordering, paper/live defaults.

Purpose

Use this skill to run a disciplined review of the current repository, not just a quick style pass.

The review should reflect the current architecture and stay aligned with docs/FINAL_SYSTEM_VISION.md (10-layer vision: notably Layer 5 = AI support, Layer 6 = learning/optimization) and enforceable rules in AGENTS.md:

• paper-only execution and position lifecycle
• hard risk gates and portfolio exposure controls
• experiment governance and scoped trials
• autonomous research loop, EXE startup path, and operator panel
• reporting, shadow analytics, and manual-review summaries
• operator observability, decision transparency, and research continuity as first-class goals

Use When

Use this skill when:

• a change spans more than one subsystem
• execution, risk, storage, analytics, or reporting were touched
• operator panel or autonomous research workflows changed
• experiment behavior, governance, or scoped-trial logic changed
• you need a findings-first review before commit or handoff

Do Not Use When

Do not use this skill for:

• trivial wording or formatting edits
• isolated markdown-only changes with no operational effect
• exploratory brainstorming before any implementation exists

Primary Workflow

1. Read AGENTS.md and, for cross-cutting changes, skim docs/FINAL_SYSTEM_VISION.md for the layers involved.
2. Identify which architecture boundaries the change crosses.
3. Review the diff with findings first, ordered by severity.
4. Check whether paper-only safety, risk ordering, and experiment isolation still hold.
5. Check whether operator-facing reporting still matches actual behavior.
6. Call out the smallest high-value next step only after findings.

Required Checks

• Confirm risk validation still sits before any new position open path.
• Confirm paper position lifecycle, sizing, and exposure reporting stay coherent.
• Confirm experiments remain opt-in and advisory where intended.
• Confirm operator panel summaries do not imply behavior the backend does not perform.
• Confirm docs, scripts, and panel actions still match the current workflow.
• Confirm no secrets, local state, or generated artifacts were accidentally pulled into tracked files.
• Confirm new autonomy or scheduling stays bounded (config/operator stops, documented limits).
• Confirm decision/score paths remain traceable (journal / decision_support / panel), not opaque.
• Confirm framing still matches the 10-layer vision, not a narrow “dashboard-only” product.

Expected Output

• Findings: prioritized issues with file references, or explicit no-findings.
• Open Questions: only unresolved assumptions that affect correctness or safety.
• Residual Risk: concise note on remaining testing or observability gaps.
• Next Step: 1 to 3 concrete follow-ups only if useful.