Arnie016/codex-prompt-templates / skills/experiment-loop-runner
experiment-loop-runner
VerifiedRun a controlled THINK-EXECUTE-REFLECT loop for coding, research, or ML experiments. Use when the user wants iterative improvement, benchmark-driven changes, autonomous experiment planning, low-cost monitoring, or progress reports without uncontrolled agent drift. Skip production release loops that require gateway policy, trace-to-regression evals, guardrail rollout, budgets, or shadow/canary traffic; use `$auto-skill-build-agent-reliability-loop`.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.0 KB |
SKILL.md
View on GitHub---
name: experiment-loop-runner
description: >
Run a controlled THINK-EXECUTE-REFLECT loop for coding, research, or ML
experiments. Use when the user wants iterative improvement, benchmark-driven
changes, autonomous experiment planning, low-cost monitoring, or progress
reports without uncontrolled agent drift. Skip production release loops that
require gateway policy, trace-to-regression evals, guardrail rollout, budgets,
or shadow/canary traffic; use `$auto-skill-build-agent-reliability-loop`.
---
# Experiment Loop Runner
Use this when progress depends on repeated trials with measurable outcomes.
## Required Control Files
```text
PROJECT_BRIEF.md
workspace/
MEMORY_LOG.md
HUMAN_DIRECTIVE.md
runs/
```
## Loop
1. **Think**: read brief, constraints, current best, and recent memory.
2. **Execute**: make one bounded change or launch one run.
3. **Verify**: run the agreed test/benchmark/dry-run.
4. **Reflect**: record result, decision, and next branch.
5. **Stop or iterate**: continue only when improvement signal exists.
## Low-Cost Monitoring
For long jobs, monitor with shell/process/log checks instead of repeatedly
asking the model to think:
- PID alive check
- tail last log lines
- GPU/CPU status when relevant
- metric parser script
## Stop Conditions
- target reached
- no improvement after N attempts
- verification becomes flaky
- human directive changes the objective
- cost/time budget exceeded
For production agent release loops that combine traces, eval groups, gateway
routing, guardrails, budgets, and shadow/canary traffic, use
`$auto-skill-build-agent-reliability-loop` first and use this skill for the
bounded experiment cycles inside that plan.
Reference pattern source: `sources/auto-deep-researcher-24x7`.
## Validation
- Each loop has a stop condition before execution.
- Each iteration records the change, verification command, result, and next decision.
- Experiments that affect production users/tools are escalated to the reliability-loop skill.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |