LLM Wiki
Harness Design for Long-Running Application Development(整理版)
- Scope: self-study
- Release: self-study-v1-e4f68c33a6f6
- Kind: topic
- Language: zh-CN
- Scope
- self-study
- Release
- self-study-v1-e4f68c33a6f6
- Source digest
- e4f68c33a6f65a13ef0ee094578a8df4bb3c033458beaa40d66c52c2593158a9
- Schema
- 1
- Published page root
- llm_wiki/published
- Manifest
- llm_wiki/reports/releases/self-study-v1-e4f68c33a6f6/release.json
- Page kind
- topic
- Page source path
- llm_wiki/published/self-study/topics/harness.md
Content
page_id: topic-harness kind: topic language: zh-CN claim_id: clm-self-study-harness-001 claim_status: synthesis
Harness Design for Long-Running Application Development(整理版)
5 分钟结构化复习
核心结论
文章关注两个相互关联的问题:
关键要点
- 文章关注两个相互关联的问题:
- 因此,迭代次数应该看质量趋势和成本收益,而不是机械地追求更多轮。
- 文章把 frontend design 中的 generator-evaluator 模式迁移到 full-stack coding,并形成三 agent 架构:
关键步骤
- 因此,迭代次数应该看质量趋势和成本收益,而不是机械地追求更多轮。
- 文章把 frontend design 中的 generator-evaluator 模式迁移到 full-stack coding,并形成三 agent 架构:
- 这个架构的关键不是“多 agent 一定更好”,而是每个 agent 都对应一个明确的模型短板。如果某个短板在新模型上已经不明显,那么对应组件就可能从必要组件变成成本负担。
自测
- 文章关注两个相互关联的问题:
场景导航
- 暂无带证据的场景关系。
来源与定位
self_study_notes/harness/整理版plan-generate-evaluate.md· line:45(claim:clm-self-study-harness-001)self_study_notes/harness/整理版plan-generate-evaluate.md· line:93(claim:clm-self-study-harness-002)self_study_notes/harness/整理版plan-generate-evaluate.md· line:97(claim:clm-self-study-harness-003)self_study_notes/harness/整理版plan-generate-evaluate.md· line:105(claim:clm-self-study-harness-004)self_study_notes/harness/整理版plan-generate-evaluate.md· line:130(claim:clm-self-study-harness-005)self_study_notes/harness/整理版plan-generate-evaluate.md· line:12(claim:clm-self-study-harness-006)
Claims
Claim metadata
-
SYNTHESIS Status meaning: Combines cited material and must not be read as a source fact.
- ID
- clm-self-study-harness-001
- Status value
- synthesis
- Source document
- self_study_notes/harness/整理版plan-generate-evaluate.md
- Locator
- line:45
-
SYNTHESIS Status meaning: Combines cited material and must not be read as a source fact.
- ID
- clm-self-study-harness-002
- Status value
- synthesis
- Source document
- Locator
-
SYNTHESIS Status meaning: Combines cited material and must not be read as a source fact.
- ID
- clm-self-study-harness-003
- Status value
- synthesis
- Source document
- Locator
-
SYNTHESIS Status meaning: Combines cited material and must not be read as a source fact.
- ID
- clm-self-study-harness-004
- Status value
- synthesis
- Source document
- Locator
-
SYNTHESIS Status meaning: Combines cited material and must not be read as a source fact.
- ID
- clm-self-study-harness-005
- Status value
- synthesis
- Source document
- Locator
-
SYNTHESIS Status meaning: Combines cited material and must not be read as a source fact.
- ID
- clm-self-study-harness-006
- Status value
- synthesis
- Source document
- Locator