autoresearch-create

name	autoresearch-create
description	Set up and run an autonomous experiment loop for any optimization target. Gathers what to optimize, then starts the loop immediately. Use when asked to "run autoresearch", "optimize X in a loop", "set up autoresearch for X", or "start experiments".

Autoresearch

Autonomous experiment loop: try ideas, keep what works, discard what doesn't, never stop.

Tools

• init_experiment — configure session (name, metric, unit, direction). Call again to re-initialize with a new baseline when the optimization target changes.
• run_experiment — runs command, times it, captures output.
• log_experiment — records result. keep auto-commits. discard/crash/checks_failed auto-reverts code changes (autoresearch files preserved). Always include secondary metrics dict. asi must always include hypothesis, rollback_reason, and next_action_hint (see below). Dashboard: command Autoresearch: Export dashboard (browser) or status bar; expand detail with Ctrl+Alt+X (VS Code/Cursor).

Setup

1. Ask (or infer): Goal, Command, Metric (+ direction), Files in scope, Constraints.
2. git checkout -b autoresearch/<goal>-<date>
3. Read the source files. Understand the workload deeply before writing anything.
4. Write autoresearch.md and autoresearch.sh (see below). Commit both.
5. init_experiment → run baseline → log_experiment → start looping immediately.

autoresearch.md

This is the heart of the session. A fresh agent with no context should be able to read this file and run the loop effectively. Invest time making it excellent.

# Autoresearch: <goal>

## Objective
<Specific description of what we're optimizing and the workload.>

## Metrics
- **Primary**: <name> (<unit>, lower/higher is better) — the optimization target
- **Secondary**: <name>, <name>, ... — independent tradeoff monitors

## How to Run
`./autoresearch.sh` — outputs `METRIC name=number` lines.

## Files in Scope
<Every file the agent may modify, with a brief note on what it does.>

## Off Limits
<What must NOT be touched.>

## Constraints
<Hard rules: tests must pass, no new deps, etc.>

## What's Been Tried
<Update this section as experiments accumulate. Note key wins, dead ends,
and architectural insights so the agent doesn't repeat failed approaches.>

Update autoresearch.md periodically — especially the "What's Been Tried" section — so resuming agents have full context.

autoresearch.sh

Bash script (set -euo pipefail) that: pre-checks fast (syntax errors in <1s), runs the benchmark, and outputs structured lines to stdout. Keep the script fast — every second is multiplied by hundreds of runs.

For fast, noisy benchmarks (< 5s), run the workload multiple times inside the script and report the median. This produces stable data points and makes the confidence score reliable from the start. Slow workloads (ML training, large builds) don't need this — single runs are fine.

Structured output

• METRIC name=value — primary metric (must match init_experiment's metric_name) and any secondary metrics. Parsed automatically by run_experiment.

Design the script to inform optimization

The script should output whatever data helps you make better decisions in the next iteration. Think about what you'll need to see after each run to know where to focus:

• Phase timings when the workload has distinct stages
• Error counts, failure categories, or test names when checks can fail in different ways
• Memory usage, cache hit rates, or other runtime diagnostics when relevant
• Anything domain-specific that would help localize regressions or identify bottlenecks

The script runs the same code every iteration — but you can update it during the loop if you discover you need more signal. Add instrumentation as you learn what matters.

Required asi keys (every log_experiment)

The VS Code/Cursor dashboard maps three columns from fixed asi keys. Every log_experiment call must pass a non-empty string for each:

Key	Meaning
hypothesis	What you believe this change will do to the primary metric and why (one or two sentences).
rollback_reason	If status is discard, crash, or checks_failed: why the change failed or was rejected (the revert survives only here). If status is keep: a short confirmation of why you kept (e.g. primary metric improved vs baseline; checks passed).
next_action_hint	The single most useful follow-up for the next iteration (or a literal none if there is nothing left to try).

Example:

"asi": {
  "hypothesis": "Caching parsed config avoids re-read on each iteration; should lower total_µs.",
  "rollback_reason": "discard: median regressed 4% vs baseline; hypothesis wrong for this workload.",
  "next_action_hint": "Try lazy parse only on first access; profile hot path before more caching."
}

You may add additional asi keys for extra signal (secondary-metric notes, profiler takeaways, etc.). Do not repeat the full description or raw benchmark output — capture what a fresh agent would need after a context reset.

Annotate failures and crashes heavily in rollback_reason. Discarded and crashed runs are reverted — the code changes are gone. The only durable record is description plus asi in autoresearch.jsonl.

autoresearch.config.json (optional)

JSON config file that lives in the pi session's working directory (ctx.cwd). Supported fields:

• maxIterations (number) — maximum experiments before auto-stopping.
• workingDir (string) — override the directory for all autoresearch operations: file I/O (autoresearch.jsonl, autoresearch.md, autoresearch.sh, autoresearch.checks.sh, autoresearch.ideas.md), command execution, and git operations. Supports absolute paths or relative paths (resolved against ctx.cwd). The config file itself always stays in ctx.cwd. Fails if the directory doesn't exist.

{
  "workingDir": "/path/to/project",
  "maxIterations": 50
}

autoresearch.checks.sh (optional)

Bash script (set -euo pipefail) for backpressure/correctness checks: tests, types, lint, etc. Only create this file when the user's constraints require correctness validation (e.g., "tests must pass", "types must check").

When this file exists:

• Runs automatically after every passing benchmark in run_experiment.
• If checks fail, run_experiment reports it clearly — log as checks_failed.
• Its execution time does NOT affect the primary metric.
• You cannot keep a result when checks have failed.
• Has a separate timeout (default 300s, configurable via checks_timeout_seconds).

When this file does not exist, everything behaves exactly as before — no changes to the loop.

Keep output minimal. Only the last 80 lines of checks output are fed back to the agent on failure. Suppress verbose progress/success output and let only errors through. This keeps context lean and helps the agent pinpoint what broke.

#!/bin/bash
set -euo pipefail
# Example: run tests and typecheck — suppress success output, only show errors
pnpm test --run --reporter=dot 2>&1 | tail -50
pnpm typecheck 2>&1 | grep -i error || true

Loop Rules

LOOP FOREVER. Never ask "should I continue?" — the user expects autonomous work.

• Primary metric is king. Improved → keep. Worse/equal → discard. Secondary metrics rarely affect this.
• Every log_experiment must include asi with hypothesis, rollback_reason, and next_action_hint (non-empty strings). Optionally add more keys. This feeds the dashboard and makes resumes cheap.
• Watch the confidence score. After 3+ runs, log_experiment reports a confidence score (best improvement as a multiple of the session noise floor). ≥2.0× means the improvement is likely real. <1.0× means it's within noise — consider re-running to confirm before keeping. The score is advisory — it never auto-discards.
• Simpler is better. Removing code for equal perf = keep. Ugly complexity for tiny gain = probably discard.
• Don't thrash. Repeatedly reverting the same idea? Try something structurally different.
• Crashes: fix if trivial, otherwise log and move on. Don't over-invest.
• Think longer when stuck. Re-read source files, study the profiling data, reason about what the CPU is actually doing. The best ideas come from deep understanding, not from trying random variations.
• Resuming: if autoresearch.md exists, read it + git log, continue looping.

NEVER STOP. The user may be away for hours. Keep going until interrupted.

Ideas Backlog

When you discover complex but promising optimizations that you won't pursue right now, append them as bullets to autoresearch.ideas.md. Don't let good ideas get lost.

On resume (context limit, crash), check autoresearch.ideas.md — prune stale/tried entries, experiment with the rest. When all paths are exhausted, delete the file and write a final summary.

User Messages During Experiments

If the user sends a message while an experiment is running, finish the current run_experiment + log_experiment cycle first, then incorporate their feedback in the next iteration. Don't abandon a running experiment.