loop-flywheel

Mine loop run history into regression cases and harness change proposals.

1|Updated Jun 20, 2026
One-click install
npx skills add https://github.com/SollanSystems/loop-engineer --skill loop-flywheel
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: loop-flywheel
Source: https://github.com/SollanSystems/loop-engineer/tree/main/skills/loop-flywheel
Command: npx skills add https://github.com/SollanSystems/loop-engineer --skill loop-flywheel

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill turns a loop's own run history into measurable improvement so repeated failures become regression cases, weak harnesses become proposals for change, and context pressure does not erase lessons learned.

Core Features & Use Cases

  • History mining: Reads RUNLOG entries, traces, and receipts to recover what actually happened across iterations.
  • Regression creation: Converts distinct real failures into durable eval cases so the same bug cannot silently return.
  • Harness improvement: Clusters recurring failure modes, recomputes loop health metrics, and proposes stopping-rule or rubric changes when the evidence shows drift.
  • Memory compaction: Separates short-term session summaries from long-term lessons so a loop can resume cleanly and improve over time.
  • Use case: A long-running coding agent starts claiming success too early; this Skill mines the false completions, adds regressions, and tightens the loop's proof-of-done process.

Quick Start

Ask the assistant to analyze the loop-flywheel skill's run history, create regression cases from confirmed failures, and compact the loop's memory into short-term and long-term summaries.

Frequently Asked Questions about loop-flywheel

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I convert recurring agent loop failures into regression test cases?

To convert recurring agent loop failures into regression test cases, mine RUNLOG entries, traces, and receipts to recover what happened across iterations, then transform distinct real failures into durable eval cases so the same bug cannot silently return.

What is the best way to compact agent loop memory when context pressure builds up?

To compact agent loop memory under context pressure, separate short-term session summaries from long-term lessons so the loop can resume cleanly and continue improving over time without erasing learned history.

How do I stop my coding agent from claiming success too early?

To stop a coding agent from claiming success too early, mine false completions from run history, add them as regression cases, and propose harness changes that tighten the loop's proof-of-done process based on evidence.

How does evidence-based metric recomputation work for agent workflows?

Evidence-based metric recomputation for agent workflows works by clustering recurring failure modes from accumulated run history and proposing stopping-rule or rubric changes when the evidence shows loop health scores drifting.

Can I propose harness changes for agent loops without silent edits?

Yes, you can propose harness changes for agent loops without silent edits by mining failure modes and recomputing loop health metrics to generate evidence-based proposals for stopping-rule or rubric adjustments.

When do I need to mine run history for loop engineering improvements?

You need to mine run history for loop engineering improvements when failures recur, scores drift, or context must be compacted, turning accumulated RUNLOG entries, traces, and repair records into durable regression protections.