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.
npx agentmods add skills/rexleimo/aios/skill-opt-litenpx skills add rexleimo/aios --skill skill-opt-litegit clone --depth 1 https://github.com/rexleimo/aiosWrote 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.
[](https://agentmods.dev/skills/rexleimo/aios/skill-opt-lite)<a href="https://agentmods.dev/skills/rexleimo/aios/skill-opt-lite"><img src="https://agentmods.dev/badge/skills/rexleimo/aios/skill-opt-lite.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00079 | $0.01687 |
| Opus 5 | $0.00039 | $0.00843 |
| Sonnet 5 | $0.00016 | $0.00337 |
| Haiku 4.5 | $0.00008 | $0.00169 |
Grade A, and why
skill-opt-lite 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.
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.
How it starts
The opening of the file, as written. The whole thing — 159 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SkillOpt-Lite: Agent-Native Skill Training
Working directory: any
Train your skill documents the way neural networks train weights — iterative rollout, reflection, and validation. No external API keys. You are both the worker and the optimizer.
When to Use
- You have a skill that doesn't work well and want to systematically improve it
- You want to create a new skill from scratch using data-driven iteration
- You want to know whether a skill change actually helps or hurts
MUST NOT use for:
- One-off skill fixes (just edit the skill directly)
- Skills that can't be objectively evaluated (purely subjective quality)
Quick Start
- Prepare a task set (JSON array of tasks with verifiable outcomes)
- Point this skill at your draft skill document
- Run the training loop
- Get an optimized
best_skill.md
Training Loop
for epoch in 1..N:
for step in 1..steps_per_epoch:
① ROLLOUT — run tasks with current skill, record pass/fail
② REFLECT — analyze failures, propose edits (≤ edit_budget)
③ AGGREGATE — deduplicate, failure-first merge
④ SELECT — pick top-L edits by impact
⑤ UPDATE — apply edits to skill document
⑥ GATE — re-run validation, accept only if score improves
SLOW_UPDATE — epoch-end strategic review into protected region
Read references/training-protocol.md for the full detailed protocol before starting a training run.
Required Inputs
| Input | Description | Format |
|---|---|---|
| skill_path | Path to the skill document to optimize | .md file |
| tasks | Task set with verifiable outcomes | JSON array (see below) |
| valid_tasks | Validation tasks (separate from training) | JSON array (optional, auto-split if not provided) |
Task format:
[
{
"id": "task-001",
"instruction": "The task the agent should perform",
"expected_outcome": "What success looks like (for scoring)",
"test_command": "optional: shell command to verify success"
}
]
Configuration Defaults
What ships with it
3 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.
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.
- 4d ago First seen · 159 lines · 79 tokens per session scan A f78fa54d923c
skill-opt-lite is a skill published in the GitHub repository rexleimo/aios (52 stars, last pushed today), licensed MIT. It adds 79 tokens to every session and 1,687 once invoked, about $0.0004 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.
Other skills, from other repositories
ml-training-recipes
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning…
optim-agent
Use when the user wants to optimize configurable system parameters against a measurable scalar objective, especially for model training, inference, quantitative strategies, reinforcement learning, scientific workflows, or other expensive black-box evaluations where reading the project can improve trial selection.
tao-run-automl
Run container-backed AutoML / hyperparameter optimization (HPO) for NVIDIA TAO networks using AutoMLRunner. Handles algorithm selection (bayesian, hyperband, asha, bohb, llm, hybrid, autoresearch), WandB experiment tracking, job execution on any TAO SDK platform, result interpretation, and per-rec custom evaluation…
training-hub-guide
Guides users through LLM post-training with Training Hub, including installation, algorithm selection (SFT, OSFT, LoRA), hyperparameter tuning, troubleshooting OOM errors, interpreting loss curves, and leveraging backend-specific features. Use when the user is working with traininghub, fine-tuning language models…
setup-guide
Use when the user wants to set up LLM training for the first time, or when traininghub is not yet installed/configured in the current environment.
memory-estimation
Use when the user wants to estimate GPU memory (VRAM) requirements for a training configuration, check if a model will fit on their GPUs, or plan GPU allocation for training.