Claude skill for minimize-ad-bug

[penelopeysm]

Closes #1313

I haven't actually tried this out yet, but I spent a bit of time iterating on it and dumping (some of) my own knowledge and experience into it. I don't think it's very complete yet. Most likely, I need to take it on a spin a few times, and then iteratively improve it.

[github-actions]

Benchmark Report

• this PR's head: cb3bc37c900ef1bc1e7ac5c2eb59d3be8da19fb8
• base branch: cca6402b74713f2a43274cc4f918dc7b8267ed2f

Computer Information

Julia Version 1.11.9
Commit 53a02c0720c (2026-02-06 00:27 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

┌───────────────────────┬───────┬─────────────┬────────┬───────────────────────────────┬────────────────────────────┬─────────────────────────────────┐
│                       │       │             │        │       t(eval) / t(ref)        │     t(grad) / t(eval)      │        t(grad) / t(ref)         │
│                       │       │             │        │ ─────────┬──────────┬──────── │ ───────┬─────────┬──────── │ ──────────┬───────────┬──────── │
│                 Model │   Dim │  AD Backend │ Linked │     base │  this PR │ speedup │   base │ this PR │ speedup │      base │   this PR │ speedup │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│               Dynamic │    10 │    mooncake │   true │   284.81 │   278.16 │    1.02 │   7.60 │    7.30 │    1.04 │   2163.37 │   2029.72 │    1.07 │
│                   LDA │    12 │ reversediff │   true │  3278.91 │  2507.07 │    1.31 │   5.02 │    2.72 │    1.85 │  16456.24 │   6811.09 │    2.42 │
│   Loop univariate 10k │ 10000 │    mooncake │   true │ 30184.10 │ 30132.68 │    1.00 │   6.81 │    6.83 │    1.00 │ 205654.28 │ 205682.52 │    1.00 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│    Loop univariate 1k │  1000 │    mooncake │   true │  3061.83 │  4098.76 │    0.75 │   6.63 │    5.03 │    1.32 │  20302.96 │  20635.86 │    0.98 │
│      Multivariate 10k │ 10000 │    mooncake │   true │ 30915.83 │ 43558.09 │    0.71 │   9.88 │    6.64 │    1.49 │ 305487.96 │ 289096.43 │    1.06 │
│       Multivariate 1k │  1000 │    mooncake │   true │  3339.31 │  3750.15 │    0.89 │   9.27 │    8.04 │    1.15 │  30950.27 │  30146.65 │    1.03 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│ Simple assume observe │     1 │ forwarddiff │  false │     0.93 │     0.84 │    1.11 │   9.58 │   11.39 │    0.84 │      8.91 │      9.53 │    0.93 │
│           Smorgasbord │   201 │ forwarddiff │  false │   914.78 │   919.25 │    1.00 │  73.89 │   74.60 │    0.99 │  67593.04 │  68578.13 │    0.99 │
│           Smorgasbord │   201 │      enzyme │   true │  1254.91 │  1252.39 │    1.00 │   4.85 │    4.43 │    1.10 │   6087.64 │   5545.69 │    1.10 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│           Smorgasbord │   201 │ forwarddiff │   true │  1255.39 │  1247.66 │    1.01 │  69.62 │   70.23 │    0.99 │  87397.34 │  87628.56 │    1.00 │
│           Smorgasbord │   201 │    mooncake │   true │  1649.04 │  1689.66 │    0.98 │   4.13 │    3.28 │    1.26 │   6807.90 │   5548.43 │    1.23 │
│           Smorgasbord │   201 │ reversediff │   true │  1242.43 │  1233.62 │    1.01 │ 127.12 │  120.57 │    1.05 │ 157942.45 │ 148735.45 │    1.06 │
├───────────────────────┼───────┼─────────────┼────────┼──────────┼──────────┼─────────┼────────┼─────────┼─────────┼───────────┼───────────┼─────────┤
│              Submodel │     1 │    mooncake │   true │     0.89 │     0.84 │    1.06 │  25.57 │   29.63 │    0.86 │     22.65 │     24.81 │    0.91 │
└───────────────────────┴───────┴─────────────┴────────┴──────────┴──────────┴─────────┴────────┴─────────┴─────────┴───────────┴───────────┴─────────┘

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.63%. Comparing base (cca6402) to head (cb3bc37).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1314   +/-   ##
=======================================
  Coverage   78.63%   78.63%           
=======================================
  Files          49       49           
  Lines        3492     3492           
=======================================
  Hits         2746     2746           
  Misses        746      746           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.