When a new model is released

Anthropic released Fable 5 this week. Here's what I'm doing with it.

Using this prompt for Fable 5:

You are auditing this codebase as a newly assigned staff engineer with no prior context. Do not assume existing patterns are intentional or correct.

Phase 1 — Orient (read-only):
- Build a mental model: architecture, module boundaries, data flow, entry points, build/test setup. Summarize in ~15 lines.

Phase 2 — Audit. For each category, report findings with:
file:line evidence, severity (critical/high/medium/low), and why it matters.
- Correctness: bugs, race conditions, resource leaks, broken error handling
- Security: injection, secrets in code, unsafe deserialization, authz gaps
- Dependency risk: outdated/vulnerable/unused dependencies, version conflicts
- Architecture: layering violations, god classes, hidden coupling,
  transaction boundary problems
- Test gaps: critical paths with no coverage, tests that assert nothing
- Dead code and drift: unused code, config that no longer matches reality, docs/comments that lie

Phase 3 — Prioritize:
- Top 10 issues ranked by (impact × likelihood) / effort
- Explicitly list 3 things you were tempted to flag but decided are actually fine, and why
- List what you could NOT verify and would need to run/test to confirm

Rules: No speculation without file:line evidence. No style nitpicks unless they cause bugs. If a finding depends on runtime behavior you can't see, say so rather than asserting it.

It will produce a detailed audit. Read it through so you understand it (it may not be perfect!). Then let Fable 5 apply the changes as you see fit.

After that, your project should be more robust and ready to work efficiently with the latest model.