minutes profile：會議記錄的 ensemble 審閱（fidelity／completeness／attribution／cross-document） - #32
Merged
Conversation
會議記錄的審閱需求既有 profile 都不合：academic 的四個 lens 有三個空轉或 方向錯誤（methodology 不適用、reference-verifier 會去查 Zotero、number-verifier 設計成跑 R/Python 重算但記錄無計算 artifact）；lecture 的 completeness （逐字稿覆蓋率）雖對得上，但 student-readability 不適用、content-accuracy 問的是知識正確性而非是否忠於逐字稿。 minutes profile 的四個 lens 互為補集：fidelity 問「記錄寫的逐字稿有嗎」、 completeness 問「逐字稿有的記錄漏了嗎」，兩者缺一不可；attribution 查發言與 責任歸屬的依據（語者分離常不可靠，未確認即具名是嚴重問題）；cross-document 比對來函、開會通知、前次記錄，並檢查交叉參照是否指對位置。DA 盯三種安靜的 偏移：個別發言寫成全體共識、條件句寫成確定句、會後才知道的事寫得像會中已知。 skill 一併記入兩個實際踩到的坑：args 傳字串會使 profile 解析為 undefined、 0 個 agent 被派出而 workflow 仍「成功」結束；agentModel 不給會繼承 session model，高階 session 單輪曾燒 56-109 萬 token。並載明與 sinica-admin:meeting-minutes 的分工——機械檢查歸 compile.sh，事實核對歸本 skill。
kiki830621
changed the base branch from
idd/29-lens-pack-externalization
to
main
August 4, 2026 02:16
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
為什麼需要新的 profile
會議記錄的 ensemble 審閱，既有三個 profile 都不合用：
academicmethodology不適用（記錄沒有研究設計）、reference-verifier會去查 Zotero、number-verifier設計成跑 R/Python 從計算 artifact 重算，但會議記錄沒有那種 artifactlecturecompleteness的「逐字稿覆蓋率」確實對得上，但student-readability不適用，content-accuracy問的是知識正確性而非「是否忠於逐字稿」code實際跑過一次
academic才確認這件事，不是憑空推論。四個 lens
fidelitycompletenessattributioncross-documentfidelity與completeness互為補集，一個防多寫、一個防漏寫，缺一不可。這是會議記錄與論文審查最不同之處：論文只怕寫錯，記錄還怕漏寫，而漏寫在文本上完全看不出來。DA 專門盯三種安靜的偏移：個別發言寫成全體共識、條件句寫成確定句（「如果可行就做」變成「將辦理」）、會後才知道的事寫得像會中已知。
Skill 記入的兩個實測陷阱
args必須是物件。傳字串時profile解析為undefined，harness 回unknown ensemble profile且 0 個 agent 被派出，但 workflow 仍「成功」結束，只在 findings 留一條 harness 層級 HIGH。stats.agents: 0是唯一的徵兆。agentModel必須顯式指定。不給時 agent 繼承 session main-loop model；engine 註解自己記載高階 session 單輪曾燒 56–109 萬 token 並在 session limit 撞死 lens agent。設計來源
lens 的每一條檢查項都來自實際做一份跨機構協調會議記錄時踩到的問題，不是抽象設計。例如
fidelity特別點名「把性質誤讀成物件」——實例是發言者說本案屬 top-down 推動，記錄卻寫成「另備 top-down 計畫書」，把計畫的性質誤讀為另一份文件；cross-document的交叉參照檢查則來自結構調整後編號語意漂移、而句子單獨讀完全正常的情形。依賴
Base 是
idd/29-lens-pack-externalization（PR #31），因為builtin-lenses.csv的重生依賴該分支引入的regen-builtin-lenses.sh。#31 merge 後本 PR 才可 merge，屆時 diff 會收斂為只剩本 PR 自身的變更。版本已 bump 至 2.22.0（含 marketplace.json 與 CHANGELOG）。