Files
united-workforce/.cards/frontmatter-fast-path.md
T
xiaomo cb3a4acf4d
CI / check (pull_request) Successful in 3m57s
chore: remove docs/, extract current knowledge to .cards
Remove 13 docs files (7 fully outdated @uncaged/* era, 6 superseded).
Extract 3 verified architectural facts as new .cards:

- frontmatter-fast-path: no LLM extraction, pure parse + schema validate + agent self-retry
- agent-cli-protocol: adapter output JSON via stdout, agent-owned step persistence
- status-based-moderator: pure graph lookup + mustache rendering, zero LLM cost

All 3 cards cross-checked against current source code (run.ts, evaluate.ts, frontmatter.ts).
2026-06-07 14:45:23 +00:00

948 B

title, created, source, tags, category, links
title created source tags category links
Frontmatter Fast-Path — No LLM Extraction 2026-06-07 openclaw-xiaomo
architecture
decision
architecture
deterministic-engine-uncertain-agent
dissipative-structure-token-for-entropy

uwf 的 agent 输出提取管线做了一个关键简化:完全不用 LLM 做结构化提取

流程:agent 输出 → 解析 YAML frontmatter → 校验 JSON Schema → 成功则继续,失败则让同一个 agent 在原 session 内追加轮次自修(最多 2 次)。

为什么不用单独的 LLM 提取:

  1. 原始 agent 有完整上下文(tool call 历史、任务理解),另起 LLM 只能猜
  2. 零额外 token 成本(fast-path 是纯字符串解析 + schema 校验)
  3. 重试走 continue() 而非新 session,保持对话连贯性

这是 PR #142 (ThreadReactor) 确立的模式。之前存在的 extract() LLM fallback 已成死代码。