Risk Check

Binding sources

• docs/FINAL_SYSTEM_VISION.md — Layer 7 hard gates; must never be bypassed by L5 or L6.
• AGENTS.md — mandatory validation order; live not default.

Purpose

Use this skill to confirm that safety still holds across the current system, not just inside app/risk.

In docs/FINAL_SYSTEM_VISION.md, hard gates map to Layer 7; this skill verifies they remain non-bypassable per AGENTS.md.

That includes:

• hard risk checks before position entry
• paper-only position lifecycle and portfolio exposure controls
• execution-quality gating
• AI and experiment isolation
• operator panel actions and startup flows

Use When

Use this skill when changes touch:

• app/risk, app/execution, app/config, app/main.py
• app/operator_panel.py or any action that can trigger runs or resets
• experiment flags, scorer confidence modes, or scoped-trial handling
• startup, EXE, or autonomous research helpers

Do Not Use When

Do not use this skill for:

• pure documentation changes
• isolated UI styling changes with no behavior or settings impact
• unrelated skill-documentation edits

Required Checks

• Confirm PAPER_TRADING=true remains the repo default.
• Confirm no live order path was added or exposed.
• Confirm hard risk validation still cannot be bypassed by AI, experiments, panel actions, or startup flows.
• Confirm learning does not auto-apply or alter hard caps.
• Confirm position sizing respects max risk, max positions, and exposure caps.
• Confirm scoped trial and full experiment only affect their intended paper-only comparison paths.

Expected Output

• Default Mode: pass/fail.
• Hard Gates: pass/fail with reasons.
• Exposure Controls: pass/fail with reasons.
• Experiment Constraints: pass/fail with reasons.
• Risk Verdict: safe / unsafe / needs follow-up.