[codex] prevent unsafe readiness-policy contradictions

[madara88645]

Summary

• make readiness consume the compiled IR safety policy that the compile route already passes in
• prevent high risk or otherwise-ready human_approval_required requests from being shown as Ready to compile
• preserve existing Clarify behavior for unresolved references and vague requests
• add unit and API regression coverage for destructive production-database requests

Root cause

The readiness analyzer accepted the compiled IR but ignored it. It independently inspected only a narrow set of text risk keywords, so a destructive request could have policy.risk_level=high and execution_mode=human_approval_required while readiness still returned ready.

User impact

The offline page now displays the readiness banner. This fix prevents that banner from presenting a green, safe-to-run message when the compiler's own policy requires review.

Scope

Changed:

• app/readiness/analyzer.py
• tests/test_readiness_analyzer.py
• tests/test_readiness_api.py

Not touched: environment files, secrets, auth, schemas, migrations, deploy/provider settings, LLM prompts, model parameters, response formats, or dependencies.

Validation

• focused readiness/safety set: 100 passed
• full backend suite: 1642 passed, 5 skipped
• pre-commit run --all-files: passed
• uv pip check: passed
• seven QA prompts re-run; the destructive database scenario now reports readiness=risky while existing clarify behavior remains intact

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
compiler	Ready	Preview, Comment	Jul 1, 2026 6:32am

[cursor]

PR Risk Assessment — Low

Decision: Approved (no reviewer assignment required)

Evidence (from diff only)

Area	Finding
Files changed	3 (app/readiness/analyzer.py, 2 test files)
Production logic	+24 / −2 lines in readiness analyzer
Blast radius	Readiness verdict only — aligns banner with compiled IR policy
Infra / auth / schema	None
Tests	Unit + API regression coverage added
CI	All checks green (Smoke, PR Tests, CodeQL, Snyk, Vercel)

What changed

Adds _policy_review() so the readiness analyzer respects policy.risk_level=high and execution_mode=human_approval_required from the compiled IR. Prevents the readiness banner from showing "Ready to compile" when the compiler policy already requires review.

Why Low (not Medium)

• Single-module, narrowly scoped behavioral fix
• Low blast radius; improves safety by removing a policy/readiness contradiction
• Straightforward logic with focused regression tests
• No shared infrastructure, auth, migrations, or prompt changes

Reviewers

None assigned — risk is below Medium threshold. No CODEOWNERS file in repo; 0 reviewers currently requested.

Slack

Slack notification tool not available in this automation run; summary posted here instead.

---

Automated risk assessment — conclusions derived from diff evidence only.

  

_{Sent by Cursor Automation: Assign PR reviewers}