agentleFS
Sign inStart free
voyager67voyager67/AgentFieldkit/skills/model-diff-review/SKILL.md

model-diff-review

Skill0 starsChanged 13 days agoMITView on GitHub

Compare two model checkpoints or prompt configurations using a consistent evaluation harness and error taxonomy. Use after fine-tuning, quantization, prompt edits, or inference-stack changes.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Verify artifact identity for both candidates: model revision/checksum, tokenizer, adapter/LoRA composition, quantization, inference …

    line 11
  2. Distinctive:

    Run baseline and candidate on identical examples.

    line 12
  3. Distinctive:

    Do not change test data or rubric after seeing results to favor one candidate.

    line 22
  4. Distinctive:

    Do not equate lower loss or faster tokens/sec with higher task quality.

    line 23

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Model Diff Review

Procedure

  1. Lock the comparison variables: evaluation rows, preprocessing, system prompt, chat template, decoding settings, tool access, timeouts, and scoring logic.
  2. Verify artifact identity for both candidates: model revision/checksum, tokenizer, adapter/LoRA composition, quantization, inference library, and runtime flags.
  3. Run baseline and candidate on identical examples. Randomize execution order if order effects, provider rate limits, or changing external state could bias results.
  4. Compare overall results and meaningful slices. For small datasets, include raw counts and uncertainty; avoid false precision.
  5. Bucket disagreements into clear categories: reasoning error, retrieval/tool failure, formatting, instruction following, refusal, hallucination, latency, or infrastructure error.
  6. Manually inspect representative wins, losses, and regressions. Verify that scoring normalization is not hiding wrong answers.
  7. Recommend ship, iterate, or reject based on predeclared criteria and risk tolerance.

Deliverable

A comparison table, top regression examples, notable improvements, operational trade-offs, reproducibility metadata, and a recommendation tied to the original goal.

Boundaries

  • Do not change test data or rubric after seeing results to favor one candidate.
  • Do not equate lower loss or faster tokens/sec with higher task quality.
  • Disclose unavailable providers, failed requests, and missing samples rather than silently dropping them.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free