v0.4.0 — 483 resources, clustered mega-sections, dataset discovery

Highlights since v0.3.0 (Jun 21)

• Resource list nearly doubled: 259 → 483 curated, source-audited entries; every URL verified reachable (0 broken)
• New section: Securing Unattended Loops — 30+ entries on prompt-injection defenses, memory-poisoning attacks, sandboxing, and monitoring for agents that run without a human watching
• The June-July research wave: 100+ new arXiv papers and benchmarks on long-horizon execution, agent memory, verification gates, and self-evolution, each title-verified against the abs page
• The Ralph ecosystem: 12 entries covering the technique and its major runners
• Mega-sections clustered: Coding-Agent Loop Systems, Verification And Feedback Gates, and Official Runtime Guides now group entries thematically
• Landing page: mobile navigation, WCAG AA/AAA contrast, print + no-JS fallbacks, Dataset JSON-LD for Google Dataset Search, refreshed social card
• Hugging Face mirror: task/size categories on the dataset card, 483-row viewer at https://huggingface.co/datasets/cy0307/awesome-loop-engineering
• 8 languages: all translations re-synced with primary-source starting links

Full list: https://github.com/ChaoYue0307/awesome-loop-engineering#readme

v0.3.0 - Patterns, runtimes, and the research wave

A large content and structure release since v0.2.0.

Highlights:

• 269+ curated entries across runtimes, research, verification, securing unattended loops, state/memory, orchestration, benchmarks, and critiques.
• 15 operational loop patterns, each with a schema-validated loop contract, plus a pattern matrix and a Choose Your Loop selector.
• 6 runnable loop templates (Claude Code /loop, desktop scheduled tasks, Codex automations, GitHub Agentic Workflows, shell/cron) and a runtime selection guide.
• New lifecycle diagram and a redesigned GitHub Pages landing site.
• Named loop concepts mapped: Ralph, Compound Engineering, Gas Town, Beads, dotskills (the ADE Loop), Omnigent, Eve.
• Research coverage of the harness-to-loop boundary, long-horizon benchmarks, agent memory, and verified orchestration.
• A synced Hugging Face mirror (cy0307/awesome-loop-engineering) and 8 language entry points.

This remains an early curated field guide for an emerging practice, not a finished standard.

v0.2.0 - Runnable loops, full contract library, security section

This release moves the repository from describing loops to shipping them.

• New runnable reference loop: examples/runnable/test-repair-loop.sh, a dependency-light test-repair loop with a deterministic verification gate, retry budget, durable progress log, repeat-failure detection, and escalation exits, plus a guide mapping it to the Loop Contract
• Loop contract library completed: all ten pattern-library loops now have schema-validated contracts in examples/
• New README section: Securing Unattended Loops (prompt injection, the lethal trifecta, OWASP agentic threats, sandboxing tools)
• Broader runtime coverage: GitHub Copilot coding agent, Google Jules, Cursor background agents, Devin, and Amp join Codex and Claude Code
• New entries for durable execution (Temporal, Restate, DBOS), observability (Langfuse, LangSmith, Phoenix, Braintrust, Weave), memory (Letta, Zep, LangMem), and long-horizon research (METR time horizons, Vending-Bench)
• Primary sources added, including Geoffrey Huntley's original Ralph technique
• Translation sync markers enforced in CI; scheduled link-check failures now auto-open an issue
• CITATION.cff for GitHub's cite button and docs/llms.txt for AI discoverability
• Maintainer docs moved to meta/ to keep the repository front page focused on content

v0.1.0 - Loop Engineering Field Guide

Initial field guide release for Awesome Loop Engineering.

Highlights:

• Curated resources for Loop Engineering as the layer above prompt, context, and harness engineering.
• Manifesto, taxonomy, comparison guide, anti-patterns, and sourced signals.
• Loop contract schema with validated example specs.
• Pattern library for PR babysitting, CI repair, docs drift, deploy verification, and feedback clustering.
• Community gallery and loop preview demo for contributed real-world loops.
• Quality gates for Markdown, links, labels, license text, loop contract examples, and clean commit identity.

This is an early curated map of an emerging practice, not a final standard.