estate-experiments
Skill14 starsChanged 22 days agoMITView on GitHub
Use when a recurring check, monitor, prompt or script should be improved by measurement instead of opinion — run a baseline-vs-trial experiment with a keep/discard receipt. Autoresearch-style loop for estate and client automation work.
In this repository13 other files it gives its agents.
- agent-reachSkill
- anthropic-skills-indexSkill
- cavemanSkill
- client-literature-monitorSkill
- cybersecurity-skills-indexSkill
- dev-pairSkill
- failure-is-not-an-optionSkill
- llm-councilSkill
- make-interfaces-feel-betterSkill
- pdf-creatorSkill
- quality-guardSkill
- senior-dev-promptSkill
- verify-resultsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 66
If the metric can't be parsed, the verdict is discard — an unmeasured improvement is not an improvement.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
estate-experiments — measure, then change
Karpathy's autoresearch pattern applied to operational work: a change to anything that runs repeatedly (a check, monitor, digest prompt, script) is proposed as an experiment, measured against the current behaviour on the same inputs, and kept only with a receipt. Opinions don't ship; receipts do.
When to Use
- You are about to reword a monitor/prompt, tune a threshold, or "optimise" a script that runs on a schedule.
- You want to know whether a change actually made things better, with evidence a reviewer can read later.
- Weekly review asks "what did we try, what did we keep".
The harness
One file, stdlib only, ships with this skill:
python3 scripts/experiment_harness.py run <experiment_dir>An experiment directory contains:
| File | Role |
intent.md | what the experiment tries to improve, and the metric (lower = better) |
baseline.sh | produces the metric on current code — read-only |
trial.sh | produces the metric with the candidate change applied |
metric.py | parses an arm's raw output into a float (parse(run_output)) |
receipt.json | written by the harness: verdict keep / discard, both metrics, why |
Rules the harness enforces:
- Both arms run against the same inputs; a missing arm is refused.
- Verdict is
keeponly if the trial is strictly better AND exits clean. - Receipts accumulate in the experiment dir; nothing is auto-applied — the human keeps the apply decision, and the receipt is the review artifact.
Worked example (real receipt)
experiments/01-dr-backup-manifest-cache/ ships as the reference: caching the morning
check's USB-manifest verification by content hash — baseline 5.21s → trial 0.19s
(96% faster), verdict keep, then applied. Copy that directory's shape for new
experiments.
Judgement
- One variable per experiment; if two things change, the receipt can't attribute.
- Warm/cold honesty: an arm that runs twice (cold then warm cache) reports the LAST timing, so a cache experiment can't cherry-pick its cold run.
- If the metric can't be parsed, the verdict is
discard— an unmeasured improvement is not an improvement. - Weekly review reads
receiptsas the experiment log; link them from your report.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.