local-distillation-lab

local-distillation-lab is a skill for Claude Code, Codex from understudylabs/understudy-agent-tools. It costs 92 tokens per session (1,915 once invoked), scanned A, original, MIT.

A local training lab for testing whether changing a small open model's learned weights improves a captured task. It compares several training approaches on a Mac using fixed training, development, and holdout data.

In plain words
What is it for?
Comparing rejection-sampled fine-tuning and distillation methods, training a local model, and measuring results with a verifiable score.
Why use it?
It helps determine whether training can improve a task after prompt changes have stopped helping, without paying for hosted training.

Skill for Claude CodeCodex

Part of the understudy plugin — 43 skills, 1 command shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/understudylabs/understudy-agent-tools/local-distillation-lab
Any agent
npx skills add understudylabs/understudy-agent-tools --skill local-distillation-lab
Clone the repo
git clone --depth 1 https://github.com/understudylabs/understudy-agent-tools

Made for: Claude Code, Codex.

Or install understudy, the plugin that ships this one along with the rest of its 43 skills, 1 command.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for local-distillation-lab

README.md
[![agentmods](https://agentmods.dev/badge/skills/understudylabs/understudy-agent-tools/local-distillation-lab.svg)](https://agentmods.dev/skills/understudylabs/understudy-agent-tools/local-distillation-lab)
Your own site
<a href="https://agentmods.dev/skills/understudylabs/understudy-agent-tools/local-distillation-lab"><img src="https://agentmods.dev/badge/skills/understudylabs/understudy-agent-tools/local-distillation-lab.svg" alt="Measured on agentmods" height="20"></a>
Per session 92 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,915 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00092 $0.01915
Opus 5 $0.00046 $0.00958
Sonnet 5 $0.00018 $0.00383
Haiku 4.5 $0.00009 $0.00192

Measured 4d ago against content hash f2d12a463979, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

local-distillation-lab scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

The scan reads SKILL.md. This mod also ships 3 executable files (examples/bench8h.py, examples/forced_likelihood.py, examples/kernel2_mlxvlm.py), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/local-distillation-lab/SKILL.md · 119 lines

How it starts

The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Local Distillation Lab

Train a local student model and measure which post-training method actually moves a captured workload — without spending on hosted training. The student samples/learns on your Mac; the only optional spend is a frontier teacher you can almost always avoid.

This is the missing weight-update rung: optimize-workload (GEPA) explicitly does not train; prepare-verifier-handoff jumps to hosted RL. This skill is what sits between them.

When To Use

  • The developer has a captured workload with a verifiable reward (final-state validator, recall/precision-vs-gold, etc.) and frozen train/dev/holdout splits.
  • Prompt optimization has plateaued and the question is now "can a weight update close the gap."
  • They want a method comparison (SFT-RS vs off-policy distill vs pedagogical/OPSD), not a single blind training run.

If the workload only needs prompt/route changes, use ../optimize-workload/SKILL.md. If it genuinely needs hosted multi-step RL, use ../prepare-verifier-handoff/SKILL.md.

Safety Gates

  • Local-first: rollouts, training, and eval run on-device. The only network call is an optional teacher; prefer the privileged self-teacher (student + gold/ICL) so nothing leaves the box.
  • Get explicit approval before any model download, frontier-teacher spend, or hosted handoff.
  • Never claim a win on oracle-tool / answer-leaked settings (e.g. AutomationBench limited_zapier hands the model the gold tools). Report the realistic-setting number alongside.
  • Holdout is sealed until a candidate adapter is frozen.

The method taxonomy (name the arms)

Dense-biased post-training methods differ by where the bias points and how concentrated it is (the concentration axis is what causes collapse):

Arm What Bias Note
B baseline, no train the floor
S rejection-sampling SFT (STaR) toward student's own passes shifts curve up, same ceiling
O off-policy distillation toward a same-family teacher's completions recipe-matched = cheap signal
P pedagogical / OPSD-style toward teacher, surprisal-gated down-weights unlearnable tokens

Read the full file on GitHub · 119 lines

Files

What ships with it

4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 119 lines · 92 tokens per session scan A f2d12a463979

Subscribe to this mod's changes

local-distillation-lab is a skill published in the GitHub repository understudylabs/understudy-agent-tools (16 stars, last pushed 3d ago), licensed MIT. It adds 92 tokens to every session and 1,915 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.

obra/superpowers · 21 tokens

brainstorming

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

obra/superpowers · 37 tokens

auto-perf-optimize

Run agent-driven VS Code performance or memory investigations. Use when asked to launch Code OSS, automate a VS Code scenario, run the Chat memory smoke runner, capture renderer heap snapshots, take workflow screenshots, compare run summaries, or drive a repeatable scenario before heap-snapshot analysis.

microsoft/vscode · 62 tokens

chat-perf

Run chat perf benchmarks and memory leak checks against the local dev build or any published VS Code version. Use when investigating chat rendering regressions, validating perf-sensitive changes to chat UI, or checking for memory leaks in the chat response pipeline.

microsoft/vscode · 51 tokens

chat-pet-sprite-creation

Use when creating or changing VS Code chat pet sprite art, sprite sheets, state animations, eye treatments, Stable/Insiders variants, or pet transitions under src/vs/workbench/contrib/chat/browser/widget/media/chatPet.

microsoft/vscode · 53 tokens

cpu-profile-analysis

Analyze V8/Chrome CPU profiles (.cpuprofile) and DevTools trace files (Trace-.json). Use when: profiling performance, investigating slow functions, comparing code paths, finding bottlenecks, analyzing timeToRequest, understanding call trees from sampling profiler data, analyzing layout/paint/rendering, investigating…

microsoft/vscode · 71 tokens