skills.rest
kodingvibes
kodingvibesOFFICIAL
Explore all skills by this author →
WebsiteGitHub

loop-architect

Designs agent loops with typed verification, cross-model review gates, and portable YAML artifacts.

2|10|Updated Jul 28, 2026
Search Tags:#verification#llm-as-judge#yaml#agent loop#loop design#kiro cli#cross-model review
One-click install
npx skills add https://github.com/kodingvibes/gamejam-2026 --skill loop-architect-kodingvibes
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: loop-architect
Source: https://github.com/kodingvibes/gamejam-2026/tree/main/participantes/jpyunism/.agents/skills/loop-architect
Command: npx skills add https://github.com/kodingvibes/gamejam-2026 --skill loop-architect-kodingvibes

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires PyYAML, and includes scripts (resource) and references (resource) components.

What problem does it solve? Designing an autonomous agent loop that actually terminates, verifies its own work, and gets reviewed by a different model is hard to get right on the first try. This Skill coaches you through goal refinement, verification criteria, reviewer/judge selection, and termination guards before you run anything, then emits runnable artifacts. ## Core Features & Use Cases - Coached Loop Design: A seven-stage interview critiques your goal, verification, council, and control settings against built-in best-practice rubrics. - Cross-Model Review Gates: Wires in reviewer and judge roles from different model CLIs (Kiro, Claude, Codex, Gemini, Ollama) with privacy redaction and consent gating. - Portable Artifacts: Emits loop.yaml, loop.resolved.json, LOOP.md, RUN_IN_SESSION.md, and a Python runner so the loop can run in-session via /goal, through subagents, or externally. - Use Case: You want an agent to repeatedly draft and revise a workflow map until a Claude judge approves it, with a 12-iteration cap and a $5 budget. The Skill designs the gates, compiles the spec, and hands off a ready-to-run prompt. ## Quick Start Ask the assistant to design an agent loop for your task using the loop-architect skill and answer its interview questions about goal, verification, and reviewers.

Frequently Asked Questions about loop-architect

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

FAQPage Schema
How do I design an agent loop with cross-model review?▼

Run the loop-architect interview, which walks through goal, verification, host model, council, and control stages. It emits a loop.yaml spec plus a RUN_IN_SESSION.md handoff prompt you can execute immediately with /goal or a subagent pipeline.

What is the difference between a reviewer and a judge in an LLM loop?▼

A reviewer writes notes only and cannot declare a gate clean, while a judge returns a structured JSON verdict of pass or revise. Gates using the revise_until_clean policy must name a judge member or human as their verdict source.

Which model CLIs can act as reviewers or judges?▼

The detection script probes for kiro-cli, claude, codex, gemini, llm, and ollama, recording their invoke argv and capabilities. Ollama is surfaced as the local, privacy-preserving option when installed.

Does loop-architect send my code to external model vendors?▼

Cross-vendor council members only receive scoped context after explicit first-send consent, with default redaction globs for .env files, secrets directories, and key files. The Skill never writes API keys or credentials into emitted artifacts.

How do I stop an agent loop from running forever?▼

The Skill requires multiple termination guards: max_iterations, a revision cap per gate, a no-progress detector that stops after repeated identical blockers, and a wall-clock or budget cap enforced by the Python runner.

When should I use the Python runner instead of running in-session?▼

Use run-loop.py for scheduled runs, CI integration, or strict wall-clock budget enforcement outside the chat session. For simple loops, following RUN_IN_SESSION.md or a /goal one-liner inside the session is faster and zero-config.

Related Skills

Explore More Software Engineering Skills→
qwibitaiqwibitai
10

add-ollama-tool

Expose local Ollama models as MCP server tools for summarization and queries.

Official
Intermediate
RohithdgrrRohithdgrr

agent-harness

Compile goals into verifiable task plans and drive bounded execute-verify-close agent loops.

Community
Advanced
rajveersingh11rajveersingh11

loop-budget

Monitor token spend against daily caps and exit loops early.

Community
Intermediate
TeiNamTeiNam
12

agentic-loops

Engineer autonomous loops with goal completion contracts and verification gates in Kiro.

Community
Advanced