experiment-runner

An autonomous assistant for running machine-learning training jobs on Databricks, a cloud platform for data and model workloads.

In plain words
What is it for?
Smoke-testing training code, submitting jobs to Databricks GPU clusters, checking warnings and failures, pulling results, and updating experiment history.
Why use it?
It handles the packaging, upload, submission, result checking, and log review needed to run experiments remotely.

Agent

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 agents/duonginspace/claude-code-databricks-ml/experiment-runner
Clone the repo
git clone --depth 1 https://github.com/duonginspace/claude-code-databricks-ml
Per session 32 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 363 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 $0.00032 $0.00363
Opus 5 $0.00016 $0.00181
Sonnet 5 $0.00006 $0.00073
Haiku 4.5 $0.00003 $0.00036

Measured 2d ago against content hash 71bec4bd1376, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

experiment-runner 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 2d 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.

agents/experiment-runner.md · 20 lines

What it actually says

You run ML experiments on Databricks GPU clusters. Always smoke-test locally before submitting. Read CLAUDE.md for the cluster config.

The submit script (scripts/submit_to_databricks.py) builds a wheel (uploaded to /mnt/dev-raw/<project-name>/ on DBFS) and uploads the training script to DBFS, and passes --wheel-path and --experiment args. The training script pip-installs the wheel at startup because DBR 15+ does not support DBFS library installs. Data/artifact files must go to /mnt/dev-raw/<project-name>/; scripts (.py, .ipynb) can go anywhere on DBFS.

After every run, read mlflow_results/job_logs.txt for the full output — even successful runs may have warnings worth noting. After every run, update mlflow_results/run_history.md with a one-line summary of the run. Never delete or overwrite mlflow_results/all_runs.csv — always append.

Common DBR 15+ failure patterns to watch for in logs:

  • pydantic has no model_validator or cannot import Sentinel from typing_extensions — stale system packages not cleared
  • BAD_REQUEST: For input string: "None" — MLflow experiment name must be /Users/... path
  • OSError: Operation not supported — script uploaded to Workspace instead of DBFS
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. 2d ago First seen · 20 lines · 32 tokens per session scan A 71bec4bd1376

Subscribe to this mod's changes

experiment-runner is an agent published in the GitHub repository duonginspace/claude-code-databricks-ml (5 stars, last pushed 5mo ago), licensed MIT. It adds 32 tokens to every session and 363 once invoked, about $0.0002 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.