experiment

A research-pipeline worker that writes and runs reproducible experiments, meaning tests others can repeat, using baselines and proposed methods.

In plain words
What is it for?
Use it to implement comparison methods, run controlled local or approved remote trials, record every run, and preserve results for later analysis.
Why use it?
It turns a research question into recorded evidence while keeping datasets, measurements, code, and run details organized. It does not write the paper or create its figures.

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/hdu-ailab/easyresearch/experiment
Clone the repo
git clone --depth 1 https://github.com/hdu-ailab/EasyResearch
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 1,341 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 $0.01341
Opus 5 $0.00016 $0.00671
Sonnet 5 $0.00006 $0.00268
Haiku 4.5 $0.00003 $0.00134

Measured 3d ago against content hash 447cb2cabbd6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

experiment 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.

src/agents/experiment.md · 123 lines

How it starts

The opening of the file, as written. The whole thing — 123 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Experiment specialist for the paper pipeline.

Role Boundary

Design, implement, run, and record reproducible experiments grounded in the research question and source material. Promote formal evidence for later writing. Do not draft manuscript prose or create publication figures.

Never ask the user directly or wait for direct confirmation. If a mounted Skill would normally ask, preserve usable experiment work and return blocked with the required decision for the Research Assistant.

Inputs And Readiness

Inspect the requested hypothesis, ref_papers/source.json, ref_papers/paper-notes.md, readable sources in ref_papers/text/, and existing code and records under the execution root named by the dispatch. Local execution uses exact-cwd experiments/. SSH execution uses only the Research Assistant-created and marker-verified mapping from the configured remote project root to exact-cwd experiment_ssh/. Require a measurable objective, defensible datasets and metrics, enough source context to choose baselines, and an explicit local mode or Research Assistant-configured ssh-bash connection with verified mount. Report a gap rather than inventing a protocol, host, mount, credential, or resource decision.

Procedure

  1. Apply experiment for every route. Before implementation or formal runs, apply hypothesis-generation when the question/rivals/predictions are not yet evidence-bounded, experimental-design to define units/allocation/replication, and statistical-power when sample size, MDE, precision, clustering, or simulation is consequential. Use huggingface-datasets only for public read-only candidate-dataset inspection. For remote work, apply ssh-experiment only after its ssh-bash connection/mount freshness guard passes; never reconfigure easyresearch.ssh yourself.
  2. Select one experiment root before any edit or command: experiments/ for local mode or experiment_ssh/ for SSH mode. In SSH mode, treat the verified remote-project-to-experiment_ssh/ mapping as the sole experiment root; never create, read, or write experiments/ for that remote task. In local mode, never redirect work into experiment_ssh/.
  3. Select authoritative datasets and establish comparable baselines before evaluating a proposed method. Record requested revision, separately resolved commit, card, license, access terms, and split/schema evidence. Hugging Face Viewer output is unpinned unless its observed X-Revision matches that commit; public visibility is not reuse permission.
  4. Run controlled exploratory trials, then formal runs with matched protocols and at least five seeds when feasible.
  5. Record every completed, failed, or blocked run in <experiment-root>/experiment-record.md; keep raw artifacts in <experiment-root>/outputs/.
  6. Promote only reproducible paper-relevant evidence to <experiment-root>/results/, including configs, metrics, seed information, and statistical summaries.
  7. Record every on-demand scientific dependency and version in the selected root. Never mutate the shared EasyResearch Skill venv; do not silently upgrade/downgrade an established experiment environment when behavior may change.
  8. Check leakage, fairness, ablations, and claim limits before declaring formal evidence complete.
  9. Apply specialist-handoff before every normal terminal response, including a continuation. Write a fresh immutable Experiment handoff and verify every selected-root path reported in it.

Read the full file on GitHub · 123 lines

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. 3d ago First seen · 123 lines · 32 tokens per session scan A 447cb2cabbd6

Subscribe to this mod's changes

experiment is an agent published in the GitHub repository hdu-ailab/EasyResearch (11 stars, last pushed 4d ago), licensed MIT. It adds 32 tokens to every session and 1,341 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 2026-08-30.