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 instructions/rchaz/tunelab/agents-mdgit clone --depth 1 https://github.com/rchaz/tunelabWrote 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/instructions/rchaz/tunelab/agents-md)<a href="https://agentmods.dev/instructions/rchaz/tunelab/agents-md"><img src="https://agentmods.dev/badge/instructions/rchaz/tunelab/agents-md.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 | $0.01026 | $0.01026 |
| Opus 5 | $0.00513 | $0.00513 |
| Sonnet 5 | $0.00205 | $0.00205 |
| Haiku 4.5 | $0.00103 | $0.00103 |
Grade A, and why
tunelab AGENTS.md 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 3d 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 — 82 lines — stays where its author put it; the contents beside it link to each section on GitHub.
AGENTS.md — running tunelab from any agent
tunelab ships as a Claude Code plugin, but the skills follow the Agent Skills
convention, so any agent (Gemini CLI, Codex, OpenClaw, or your own harness) can drive
the same workflow. This file is the entry point: read it, then read the relevant
skills/<name>/SKILL.md for the step you're on.
What tunelab does
Helps a user move repetitive LLM work — classifying, routing, extracting, drafting — onto small/local models, and proves the cheaper option works on their data before they ship it. The core idea is a capability ladder of numbered levels (Level -1 through 3): try the cheapest level that clears the user's accuracy bar, and only escalate when the task demands it. Talking a user out of fine-tuning by showing a cheaper level already meets their bar is a success, not a failure.
Start here, always
Begin every task by reading skills/tune-decide/SKILL.md and following it — even if the
user has already decided to fine-tune. It interviews the user, runs cheap probes plus a
frontier "ceiling" probe on their own data, and recommends a level with evidence. It then
routes to the other skills as needed.
The skills and when each applies
| Skill | Read its SKILL.md when |
|---|---|
tune-decide |
The front door. Always run first. Decides which level fits and executes the cheap levels inline. |
tune-data |
A run needs a clean train/test split, deduping, or labels generated by a teacher model. |
tune-train |
The chosen level is a local LoRA / QLoRA / full fine-tune / continued-pretraining (MLX, Apple Silicon). |
tune-eval |
You need held-out accuracy, LLM-as-judge scoring, or cascade composition — with the bar set before scores are seen. |
tune-loop |
Champion/challenger: promote a new model only when it beats the incumbent by a pre-registered margin. |
Typical path: tune-decide → tune-data → tune-train → tune-eval, with tune-loop
wrapping the improvement cycle.
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.
- 3d ago First seen · 82 lines · 1,026 tokens per session scan A a9f2afffa32c
tunelab AGENTS.md is an instructions file published in the GitHub repository rchaz/tunelab (6 stars, last pushed 1mo ago), licensed MIT. It adds 1,026 tokens to every session, about $0.0051 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 2026-08-31.
Other instructions, from other repositories
tuning-engines-cli CLAUDE.md
Instructions for cerebrixos-org/tuning-engines-cli, covering tuning engines mcp server, setup, claude desktop configuration, claude code and available tools.
axolotl AGENTS.md
AGENTS.md instructions for axolotl-ai-cloud/axolotl, covering axolotl, tech stack, commands, training methods and config pattern.
prompty copilot-instructions.md
Instructions for microsoft/prompty, covering prompty v2 — complete rebuild plan, what prompty uses from the typra-generated model, key design decisions, python coding rules and environment & tooling.
pi-tinker AGENTS.md
Instructions for gvkhosla/pi-tinker, covering agent guide for pi-tinker, product promise, agent compatibility, best first commands for users and what is possible.
pi-tinker copilot-instructions.md
Instructions for gvkhosla/pi-tinker, a project described as: Fine-tune open-source models with Tinker from inside Pi — managed improve loops, data prep, evals, smoke tests, deploy snippets, and checkpoint chat.
ai4science-studio CLAUDE.md
Instructions for AMDResearch/ai4science-studio, covering claude.md, repository purpose, agent entry points (read these first), directory layout and model slug rule.