会议议程与行动清单

判断是否需要开会，整理带时长和目标的议程，把会议笔记转换为负责人、截止日期明确的行动项。

github最新版本 0.1.1更新于 2026-10-21下载 2 次2 个版本

references/action_item_discipline.md

1	# Action-Item Discipline — why meetings without owned actions are theater
2
3	The third discipline of the `meetings` skill: the meeting's output is not the discussion, the
4	slides, or even the decision — it is the set of committed next actions that exist afterwards. A
5	meeting that ends without owned, dated actions consumed its budget and produced a feeling. This
6	file documents the canon behind `action_item_extractor.py` and its two flags, ORPHAN (no
7	owner) and NO-DUE (no date).
8
9	## The next-action rule (why vague intentions don't survive)
10
11	David Allen's Getting Things Done (Penguin, 2001) built its entire system on one observation:
12	outcomes don't get done, next actions do. "Handle the pricing situation" is a project;
13	"Maria sends the pricing one-pager by Friday" is an action. Allen's test — *what's the very next
14	physical, visible activity, and who takes it?* — is exactly the shape the extractor demands: a
15	verb, an owner, a date. Anything in the notes that fails the test ("we should think about X",
16	"the team agreed X matters") is deliberately not extracted, because promoting sentiment to a
17	checklist launders vagueness into false progress.
18
19	## One owner, not a committee
20
21	Every extracted item gets exactly one owner or an ORPHAN flag. This is the "Driver" /
22	"Accountable" insight of the DACI and RACI decision frameworks (documented in the Atlassian Team
23	Playbook's DACI play): when a task has two owners it has zero, because each can reasonably assume
24	the other has it. The extractor never auto-assigns an orphan — assignment is a human commitment,
25	and a name typed by a script is not a commitment. The flag's job is to force the assignment
26	conversation while the room (or thread) still remembers agreeing to the work.
27
28	## A date, or it drifts
29
30	George Doran's original SMART memo ("There's a S.M.A.R.T. Way to Write Management's Goals and
31	Objectives," Management Review, 1981) made time-bound one of the five criteria for a reason:
32	an undated commitment cannot be late, so it never becomes urgent, so it loses every scheduling
33	contest against work that can. The extractor captures "by/due/before <date>" phrasings and flags
34	the rest NO-DUE. The fix costs five seconds at recap time ("Alex — by when?") and is the
35	highest-leverage edit anyone makes to meeting notes.
36
37	## Implementation intentions (the psychology of "who + when")
38
39	The owner + date rule is not managerial superstition; it has a mechanism. Peter Gollwitzer's
40	implementation-intentions research shows that goals formulated as if-then plans — specifying
41	when, where, and how the action happens — are acted on at substantially higher rates than
42	equally-motivated abstract goals, across dozens of studies (Peter M. Gollwitzer, "Implementation
43	Intentions: Strong Effects of Simple Plans," American Psychologist, 1999; meta-analysis:
44	Gollwitzer & Sheeran, Advances in Experimental Social Psychology, 2006). An action item with an
45	owner and a date is an implementation intention; one without is a wish. Likewise, Locke and
46	Latham's goal-setting research — the most replicated result in organizational psychology — shows
47	specific, difficult goals reliably outperform "do your best" vagueness (Edwin A. Locke & Gary P.
48	Latham, "Building a Practically Useful Theory of Goal Setting and Task Motivation,"
49	American Psychologist, 2002).
50
51	## Follow-through is the dropped phase
52
53	Steven Rogelberg's meeting research finds the post-meeting phase is where most meeting value
54	leaks: actions go unrecorded, unowned, or unrevisited, and recurring meetings quietly re-discuss
55	last week's conclusions (The Surprising Science of Meetings, Oxford University Press, 2019).
56	His prescriptions map one-to-one onto this skill's mechanics: end with a recap of commitments
57	(the agenda builder's mandatory closing slot), record owner + deadline for each (the extractor's
58	output shape), and open the next meeting against the last list (the emitted markdown checklist is
59	built to be pasted into the next agenda's pre-read).
60
61	## Why extraction is deterministic
62
63	The extractor uses fixed patterns — checkboxes, `ACTION:`/`TODO:` prefixes, "@name will …",
64	"Name will … by date" — rather than judgment. Two reasons. First, auditability: anyone can look at
65	a line of notes and know whether it will extract, so the discipline is teachable ("write it as
66	`ACTION: @sam … by Friday` and it will never be lost"). Second, honesty: a fuzzy extractor that
67	infers actions from vibes recreates the original problem — commitments nobody actually made. If
68	the notes contain no extractable actions, the correct output is the uncomfortable summary line
69	saying so, not an invented checklist.
70
71	## Sources
72
73	1. David Allen, Getting Things Done: The Art of Stress-Free Productivity, Penguin, 2001
74	(next-action discipline; the verb-owner test).
75	2. Steven G. Rogelberg, The Surprising Science of Meetings, Oxford University Press, 2019
76	(follow-through as the dropped phase; end-with-recap prescription).
77	3. Atlassian Team Playbook, "DACI: Decision-making framework" (single Driver/Accountable;
78	two owners = zero owners).
79	4. George T. Doran, "There's a S.M.A.R.T. Way to Write Management's Goals and Objectives,"
80	Management Review, 1981 (time-bound as a first-class criterion).
81	5. Peter M. Gollwitzer, "Implementation Intentions: Strong Effects of Simple Plans," *American
82	Psychologist*, 1999 (and Gollwitzer & Sheeran 2006 meta-analysis) — who/when/where plans get
83	acted on; wishes don't.
84	6. Edwin A. Locke & Gary P. Latham, "Building a Practically Useful Theory of Goal Setting and
85	Task Motivation," American Psychologist, 2002 (specific goals beat vague ones).

文件内容是平台检查时读出的文本，只供查看。平台不会运行包里的任何脚本或程序。