agentsclimarketplace

Experiment log report

Skill StaryMoon/experiment-log-report-skill/skills/experiment-log-report

Agent Skill for auditable experiment reports from CSV and JSONL training logs.

Install
npx -y skills add StaryMoon/experiment-log-report-skill --skill experiment-log-report

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

2 things to look at

  • 24 days oldThe repository was created 24 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Convert CSV or JSONL training logs into metric curves, best-step summaries, anomaly notes, and a reproducible Markdown report. Use when Codex needs to compare runs, summarize a long experiment, find regressions or NaNs, prepare a lab update, or turn raw machine logs into evidence a collaborator can audit.

SKILL.md

1.6 KB, 273 tokens by cl100k_base, as published. Nobody here has run it

Experiment Log Report

Build the report from raw logs, not copied terminal snippets. Preserve run names, metric directions, best steps, and source paths so every claim can be checked.

Workflow

  1. Identify CSV or JSONL files containing one record per step or epoch.
  2. Run one or several logs together:
python3 scripts/build_experiment_report.py runs/baseline.csv runs/new-method.jsonl \
  --output output/experiment-report \
  --metrics psnr,ssim,loss
  1. Inspect curves.png for sparse logging, axis-scale problems, and truncated runs.
  2. Read summary.json before writing conclusions. Metrics containing loss, error, lpips, fid, wer, or latency are minimized by default; other metrics are maximized.
  3. Add interpretation below the generated evidence in report.md. Do not rewrite the generated source table.

Quality rules

  • Compare runs only on overlapping metrics.
  • Distinguish best checkpoint from final checkpoint.
  • Surface NaN and infinite values instead of dropping them silently.
  • Do not smooth curves unless the user asks; if smoothing is added, retain raw curves.
  • Record any manual direction override in the report.

See references/log-schema.md for accepted input shapes.

Keep looking

Skills are one crate of 328,083. 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.