review-gated-loops
Design agent loops with explicit verification gates before execution and judging before delivery — Goal → Plan → Review → Deliver → Judge → Stop. Use when an agent task needs pre-flight checking, cross-model reviewer councils, or stop-conditions to avoid runaway/low-quality output. Complements the OS debate_protocol (post-output quality gate) with the pre-flight + loop-control half.
- License 许可
- MIT
- 最近更新
- 2026-10-09
- Stars
- ★ 2
- Forks
- 1
01 前言 Front Matter
- name
- review-gated-loops
- license
- MIT
- metadata
- { "type": "harness-pattern", "source": "https://github.com/ksimback/looper", "complements": "1_CORE/scripts/core/debate_protocol.py" }
- description
- Design agent loops with explicit verification gates before execution and judging before delivery — Goal → Plan → Review → Deliver → Judge → Stop. Use when an agent task needs pre-flight checking, cross-model reviewer councils, or stop-conditions to avoid runaway/low-quality output. Complements the OS debate_protocol (post-output quality gate) with the pre-flight + loop-control half.
03 SKILL.md 正文
Review-gated agent loops (looper)
ksimback/looper (MIT, Python) — a skill for structuring agent loops with verification gates: Goal → Plan → Review → Deliver → Judge → Stop. Two ideas worth adopting in the OS harness:
- Pre-flight verification gates — validate the plan before executing (cheap to catch a bad approach before spending tokens/side-effects).
- Cross-model reviewer council — have N independent reviewers (ideally different models) judge before delivery; require majority approval.
Relation to the OS's debate_protocol.py
1_CORE/scripts/core/debate_protocol.py already runs a post-output gate
(run_debate): Round 1 internal-consistency, Round 2 knowledge-alignment vs vector
memory → APPROVED / NEEDS_REVISION. That is the Judge step.
looper adds the missing halves: the pre-flight Review gate (check the plan first)
and the reviewer council (multiple independent judges, not one pass). A natural
upgrade path: wrap dispatch so a task runs plan → pre-flight gate → execute → run_debate(output) → stop-or-revise, and let run_debate call a small council of
reviewers instead of a single pass.
debate_protocol is the Judge; looper is the loop + gates around it.
04 同仓库更多 Skills
web_crawler_katana
High-speed web crawling using Katana patterns for deep site mapping, URL discovery, and content extraction.
server_log_analyzer
Analyzes Nginx/Apache server logs to identify Googlebot crawl patterns, errors, and performance bottlenecks.
schema_markup_generator
Auto-generates JSON-LD structured data for any page type (Article, Product, FAQ, LocalBusiness, etc.).
internal_linking_optimizer
Analyzes and optimizes the internal link graph of a website for SEO authority distribution.
google_trends_analyzer
Detects trending topics and seasonal patterns using Google Trends data for content planning and keyword strategy.
brand_context_analyzer
Tracks brand mentions, sentiment, and online reputation across web sources.