Summary
当需要对改进候选多人盲审打分、用 LLM 做语义评估、判断候选是否应被接受、或打分结果全是 hold 想知道为什么时使用。支持 --panel 多审阅者盲审和 --llm-judge 语义评估。不用于结构评估(用 improvement-learner)或门禁决策(用 improvement-gate)。
lanyasheng/auto-improvement-orchestrator-skill · Archived
当需要对改进候选多人盲审打分、用 LLM 做语义评估、判断候选是否应被接受、或打分结果?
npx skills add lanyasheng/auto-improvement-orchestrator-skill --skill improvement-discriminator
当需要对改进候选多人盲审打分、用 LLM 做语义评估、判断候选是否应被接受、或打分结果全是 hold 想知道为什么时使用。支持 --panel 多审阅者盲审和 --llm-judge 语义评估。不用于结构评估(用 improvement-learner)或门禁决策(用 improvement-gate)。
This repository is archived — consider an actively maintained alternative.
当需要把已批准的改进候选应用到目标文件、回滚之前的变更、或预览变更效果时使用。支持 4 种 action…
1 installs当需要为已有 Skill 自动生成 task_suite.yaml 测试任务集、从 skill_spec.yaml 生成完整 SKILL.md + …
1 installs当需要把多个功能重叠的 skill 合并为一个蒸馏版 skill 时使用。 不适用于从 skills 提取 rules(rule…
1 installsParse Claude Code session JSONL to extract implicit user feedback signals. Detects skill invoca…
1 installsRelated neighbors and high-traction skills in the same topics — useful to compare before installing.
Helps users discover and install agent skills when they ask questions like "how do I do X", "fi…
3.3M installsBrowser automation CLI for AI agents. Use when the user needs to interact with websites, includ…
810.4K installsReview UI code for Web Interface Guidelines compliance. Use when asked to "review my UI", "chec…
617.3K installsBuild, deploy, evaluate, optimize, fine-tune, and manage Microsoft Foundry agents, models, and …
576.5K installsPrepare azd-based Azure projects for deployment: generates azure.yaml, infrastructure (Bicep/Te…
568.3K installsOther skills from lanyasheng/auto-improvement-orchestrator-skill · top by installs.
npx skills add lanyasheng/auto-improvement-orchestrator-skill
Browse all from lanyasheng/auto-improvement-orchestrator-skill
Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.
main
Parsed from SKILL.md frontmatter.
Files included with this skill beyond the listing page.
SKILL.md
8,285 B
SUMMARY.md
352 B
Multi-signal scoring engine: heuristic rules + evaluator rubrics + LLM-as-Judge + multi-reviewer blind panel.
improvement-learner(learner 做 6 维结构分析,discriminator 做语义评分)improvement-gate(discriminator 只打分不做决策)improvement-executor(discriminator 不修改任何文件)improvement-generator4 种模式可以独立或组合使用,从纯启发式到全信号融合逐步增加评估深度。 默认模式(Heuristic only)零成本、确定性,适合快速过滤。 加入 evaluator evidence 后,利用 task_suite 执行结果作为评分依据,更贴近实际效果。 加入 LLM Judge 后引入语义理解,但会消耗 token 且结果有随机性。 Panel 模式引入多视角盲审,捕捉单一审阅者无法发现的偏差。
| Mode | Flag | Scoring |
|---|---|---|
| Heuristic only | (default) | category bonus + source refs + risk penalty |
| + Evaluator | --use-evaluator-evidence |
Heuristic 70% + evaluator 30% |
| + LLM Judge | --llm-judge {claude,openai,mock} |
Heuristic 60% + LLM 40% |
| + Panel | --panel |
2+ reviewers independently, cognitive label decides |
| All combined | --panel --llm-judge mock --use-evaluator-evidence |
Full |
Tradeoff: single reviewer speed vs. multi-reviewer accuracy.
之所以引入 panel 盲审而非依赖单一评分器,原因是:
问题: 为什么不直接用 LLM judge 替代所有启发式规则?Because LLM judge 有 token 成本(每个候选约 500-1000 tokens)且存在随机性。启发式规则是确定性的、零成本的,适合作为第一层过滤。组合使用时,启发式占 60% 权重、LLM 占 40%,既保证了稳定性又引入了语义理解。
当 panel 结果全是 hold 时,通常是以下原因之一:
sourcerefs(引用来源),导致 sourceref_bonus = 0risklevel 被标为 high,导致 riskpenalty 过大<example> 正确用法: 多审阅者盲审 + LLM 语义打分 $ python3 scripts/score.py --input candidates.json --panel --llm-judge mock --output scored.json → 输出包含: panelreviews: [{reviewer: "structural", score: 7.5}, {reviewer: "conservative", score: 5.0}] cognitivelabel: "VERIFIED" (2人同意) llm_verdict: {score: 0.78, decision: "conditional", dimensions: {clarity: 0.85, ...}} </example>
<anti-example> 常见误解: --panel 和 --llm-judge 互斥 → 错!两者可以同时使用。每个审阅者独立调用 LLM judge,得到独立的语义分数。 → 如果只用 --panel 不加 --llm-judge,panel 只做启发式评分,不做语义评估。 </anti-example>
score.py 是核心入口,接收 candidates.json,输出 scored.json。 所有模式共用同一个入口,通过 flag 组合控制评分深度。 --llm-judge 支持 3 种 provider: claude(最准)、openai、mock(测试用,零成本)。 --panel 会自动创建 structural 和 conservative 两个独立审阅者。 输出的 scored.json 可直接传给 executor 或 gate 消费。 使用 --verbose 可查看每个审阅者的详细评分过程和 judge_notes。
# Basic scoring (heuristic only, fastest)
python3 scripts/score.py --input candidates.json --output scored.json
# Full pipeline: panel + LLM judge
python3 scripts/score.py \
--input candidates.json --panel --llm-judge mock --output scored.json
Panel-only mode (no LLM, lower cost):
# Panel blind review without LLM judge — heuristic scoring only
python3 scripts/score.py \
--input candidates.json \
--panel \
--output scored.json
LLM-judge-only mode (no panel, single reviewer):
# Single reviewer + LLM semantic evaluation
python3 scripts/score.py \
--input candidates.json \
--llm-judge claude \
--use-evaluator-evidence \
--output scored.json
| Request | Deliverable |
|---|---|
| Score | JSON: per-candidate scores, blockers, recommendations, judge_notes |
| Panel | JSON: panelreviews[], cognitivelabel (CONSENSUS/VERIFIED/DISPUTED), aggregated_score |
| LLM judge | JSON: llm_verdict with score, decision (accept/conditional/reject), 4 dimensions, confidence |
| Combined | All above fields merged into a single scored candidate object |
输出中的 cognitive_label 含义:
decision 字段的三态语义:accept 直接通过,conditional 需要满足附加条件(记录在 judge_notes),reject 直接拒绝。
--use-evaluator-evidence 的数据源Pipeline 中的数据流: generator → discriminator → evaluator → executor → gate