agentsclimarketplace

Ml

Skill telagod/code-abyss/skills/_kernel/ml

Machine learning and LLM engineering judgment, distilled from a stronger model - invoke when DECIDING whether/how to use ML or an LLM for a task (prompt vs RAG vs fine-tune vs classical); working with training/eval data or labels; building or reviewing evals for models and LLM features; designing RAG, structured output, or agent pipelines; or diagnosing why a model/LLM feature underperforms. Method-selection ladder, data and leakage discipline, eval-as-spec rules, LLM-era craft, and a trap catalog.From its SKILL.md

Install
npx -y skills add telagod/code-abyss --skill ml

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

2.2 KB, 455 tokens by cl100k_base, as published. Nobody here has run it

ML — approach, data, evals, LLM craft, traps

Rule content lives in the five files below; this SKILL.md only routes (doctrine/04-maintenance.md governs edits to this bundle too).

Route by moment

You are about to…Read (in this folder)
Decide whether ML/an LLM is warranted, and which method rung to useapproach.md
Touch a dataset, labels, or splits; suspect a score is too gooddata.md
Define success, build/judge an eval, or assess someone's metric claimevals.md
Build with LLMs: prompts, RAG, structured output, agents, model choicellm.md
Diagnose an underperforming model or LLM featuredata.md §1 first (read real failures), then llm.md §3 if RAG, traps.md to name the pattern
Review an ML project's health; name why a claim or pipeline smells wrongtraps.md

A new ML feature usually runs approach.md (interrogate + pick the rung) → evals.md §1 (eval BEFORE build) → data.md → then llm.md if the rung is LLM-shaped → skim traps.md §C before finalizing any launch or monitoring plan.

Scope and neighbors

Modeling and evaluation judgment. The serving infrastructure around a model is ordinary backend (backend bundle: APIs, queues, operate.md); experiment execution discipline is methods (investigate/verify); whether to delegate → doctrine.

The stance

The eval is the spec; anything unmeasured is folklore. Look at the data with your own eyes (data.md §1), climb the method ladder from the cheapest rung (approach.md §3), and treat every surprising score as leakage until disproven (data.md §2). The failure mode of this field is not bad models — it is unearned confidence in numbers.

What ships with it: 5 files

19.7 KB alongside SKILL.md

Gives 0 of the 12 instructions most evals benchmarks skills give in 455 tokens

Counted across 499 of the 513 authors here whose files we hold, read 2026-09-06

  • Spawn with-skill and baseline runs in the same turnin 31 of 499, across 24 files
  • Keep SKILL.md under 500 linesin 31 of 499, across 24 files
  • Draft assertions while test runs are in progressin 31 of 499, across 24 files
  • Compare against the baseline after changesin 31 of 499, across 13 files
  • Define evals before codingin 26 of 499, across 17 files
  • Run evals frequently during developmentin 25 of 499, across 16 files
  • Keep evals fastin 24 of 499, across 15 files
  • Version evals with codein 24 of 499, across 15 files
  • Generate the eval viewer before evaluating outputs yourselfin 24 of 499, across 17 files
  • Generate an eval report after runsin 24 of 499, across 15 files
  • Track pass@k metrics over timein 22 of 499, across 14 files
  • Save a baseline before making changesin 21 of 499, across 9 files

Said here and by no other author read

  • read the rule file matching the current task
  • look at the data yourself before trusting any score
  • start from the cheapest rung of the method ladder
  • treat every surprising score as leakage until disproven
  • follow the routed reading sequence for new ML features
  • read real failures first when diagnosing underperformance

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.