opencode-swarm-plugin 0.12.25 → 0.12.26
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.beads/issues.jsonl +17 -0
- package/README.md +105 -0
- package/global-skills/agent-patterns/SKILL.md +682 -0
- package/global-skills/learning-systems/SKILL.md +644 -0
- package/global-skills/mcp-tool-authoring/SKILL.md +695 -0
- package/global-skills/resilience-patterns/SKILL.md +648 -0
- package/global-skills/tacit-knowledge-extraction/SKILL.md +387 -0
- package/global-skills/testing-strategies/SKILL.md +558 -0
- package/global-skills/zod-validation/SKILL.md +763 -0
- package/package.json +1 -1
|
@@ -0,0 +1,387 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: tacit-knowledge-extraction
|
|
3
|
+
description: Extract tacit knowledge into pattern languages. Use when creating skills from expert knowledge, documenting tribal knowledge, or turning implicit expertise into explicit patterns. Meta-skill for creating better skills.
|
|
4
|
+
tags: [knowledge-management, patterns, learning, documentation]
|
|
5
|
+
---
|
|
6
|
+
|
|
7
|
+
# Tacit Knowledge Extraction
|
|
8
|
+
|
|
9
|
+
Turn implicit expertise into explicit, shareable patterns. Extract the "tricks of the trade" that experts can't easily articulate.
|
|
10
|
+
|
|
11
|
+
## What is Tacit Knowledge
|
|
12
|
+
|
|
13
|
+
**Tacit knowledge** = knowledge that's difficult to articulate, procedural, heuristic, gained through direct experience. The "Fingerspitzengefühl" that experts have but can't explain.
|
|
14
|
+
|
|
15
|
+
**Characteristics:**
|
|
16
|
+
|
|
17
|
+
- Hard to verbalize ("I just know")
|
|
18
|
+
- Procedural, not declarative
|
|
19
|
+
- Context-dependent
|
|
20
|
+
- Gained through practice, not instruction
|
|
21
|
+
- Often unconscious competence
|
|
22
|
+
|
|
23
|
+
**Examples:**
|
|
24
|
+
|
|
25
|
+
- How a senior dev "smells" bad code
|
|
26
|
+
- When to break the rules vs follow them
|
|
27
|
+
- Debugging intuition ("check this first")
|
|
28
|
+
- Knowing when a design is "done"
|
|
29
|
+
|
|
30
|
+
**Why it matters:** Most valuable knowledge in organizations is tacit. If you don't extract it, it walks out the door when experts leave.
|
|
31
|
+
|
|
32
|
+
## Pattern Language Structure
|
|
33
|
+
|
|
34
|
+
A **pattern** describes a recurring solution to a problem in a context. Format from Christopher Alexander (architecture) → software → human activities.
|
|
35
|
+
|
|
36
|
+
### Core Elements
|
|
37
|
+
|
|
38
|
+
**Name:** Evocative, memorable (2-4 words)
|
|
39
|
+
|
|
40
|
+
- Good: "Experienced Person", "Radar Chart Review"
|
|
41
|
+
- Bad: "Knowledge Extraction Method #3"
|
|
42
|
+
|
|
43
|
+
**Context:** When does this apply?
|
|
44
|
+
|
|
45
|
+
- Situation, preconditions, scope
|
|
46
|
+
- "You are creating a skill from expert knowledge..."
|
|
47
|
+
|
|
48
|
+
**Problem:** What tension needs resolving?
|
|
49
|
+
|
|
50
|
+
- The force that creates the need
|
|
51
|
+
- "Experts can't articulate what they know..."
|
|
52
|
+
|
|
53
|
+
**Solution:** The pattern itself (imperative form)
|
|
54
|
+
|
|
55
|
+
- Actionable, specific steps
|
|
56
|
+
- "Interview 5-10 practitioners with well-balanced selection..."
|
|
57
|
+
|
|
58
|
+
**Consequences:** What results? Trade-offs?
|
|
59
|
+
|
|
60
|
+
- Benefits and liabilities
|
|
61
|
+
- "You get diverse perspectives but need more synthesis time"
|
|
62
|
+
|
|
63
|
+
**Forces:** Competing concerns the pattern balances
|
|
64
|
+
|
|
65
|
+
- Optional but valuable for complex patterns
|
|
66
|
+
- E.g., speed vs thoroughness, breadth vs depth
|
|
67
|
+
|
|
68
|
+
### Pattern Relationships
|
|
69
|
+
|
|
70
|
+
Patterns connect to form a **language**:
|
|
71
|
+
|
|
72
|
+
- **Uses:** This pattern employs another
|
|
73
|
+
- **Refines:** This pattern is a specific case of another
|
|
74
|
+
- **Precedes:** Do this before that
|
|
75
|
+
- **Alternatives:** Different solutions to same problem
|
|
76
|
+
|
|
77
|
+
## Extraction Process
|
|
78
|
+
|
|
79
|
+
Collaborative process to mine patterns from practitioners. Based on Iba et al.'s procedure.
|
|
80
|
+
|
|
81
|
+
### Step 1: Select Experienced People
|
|
82
|
+
|
|
83
|
+
**Pattern: EXPERIENCED PERSON**
|
|
84
|
+
Choose people who are well-experienced and admirable in the field you're documenting.
|
|
85
|
+
|
|
86
|
+
**Pattern: WELL-BALANCED SELECTION**
|
|
87
|
+
Select 5-10 people with diverse perspectives within the domain. Different specialties, seniorities, contexts.
|
|
88
|
+
|
|
89
|
+
**Why diversity matters:** Each person knows different aspects. Overlap reveals core patterns, gaps reveal edge cases.
|
|
90
|
+
|
|
91
|
+
### Step 2: Interview for Experiences
|
|
92
|
+
|
|
93
|
+
**Pattern: EXPERIENCE MINING**
|
|
94
|
+
Don't ask "what do you think?" Ask "tell me about a time when..."
|
|
95
|
+
|
|
96
|
+
**Interview techniques:**
|
|
97
|
+
|
|
98
|
+
- **Critical incidents:** "Tell me about a really successful project. What made it work?"
|
|
99
|
+
- **Failure analysis:** "When did it go wrong? What would you do differently?"
|
|
100
|
+
- **Contrast cases:** "How is X different from Y?"
|
|
101
|
+
- **Concrete examples:** Force specifics, not generalities
|
|
102
|
+
|
|
103
|
+
**Questions to ask:**
|
|
104
|
+
|
|
105
|
+
- "Walk me through how you do X"
|
|
106
|
+
- "What do you check first when debugging?"
|
|
107
|
+
- "How do you know when it's good enough?"
|
|
108
|
+
- "What mistakes do beginners make?"
|
|
109
|
+
- "What rules do you break, and when?"
|
|
110
|
+
|
|
111
|
+
**Observation beats interviews:**
|
|
112
|
+
If possible, watch experts work. Note what they do, not what they say they do. Experts often can't articulate unconscious moves.
|
|
113
|
+
|
|
114
|
+
### Step 3: Extract Pattern Candidates
|
|
115
|
+
|
|
116
|
+
**Pattern: PATTERN MINING**
|
|
117
|
+
After interviews, identify recurring themes across practitioners.
|
|
118
|
+
|
|
119
|
+
**Mining techniques:**
|
|
120
|
+
|
|
121
|
+
1. **Affinity mapping:** Cluster similar experiences
|
|
122
|
+
2. **Contrast analysis:** What do experts do that novices don't?
|
|
123
|
+
3. **Negative cases:** When does the pattern NOT apply?
|
|
124
|
+
4. **Threshold detection:** When does one pattern transition to another?
|
|
125
|
+
|
|
126
|
+
**Look for:**
|
|
127
|
+
|
|
128
|
+
- Repeated phrases ("I always...", "First thing I do...")
|
|
129
|
+
- Shared heuristics (rules of thumb)
|
|
130
|
+
- Common mistakes mentioned by multiple people
|
|
131
|
+
- Things experts do unconsciously
|
|
132
|
+
- Disagreements (might indicate context variations)
|
|
133
|
+
|
|
134
|
+
### Step 4: Validate with Practitioners
|
|
135
|
+
|
|
136
|
+
**Pattern: PRACTITIONER VALIDATION**
|
|
137
|
+
Test draft patterns with people who weren't interviewed. Do they recognize them? Can they use them?
|
|
138
|
+
|
|
139
|
+
**Validation methods:**
|
|
140
|
+
|
|
141
|
+
- **Recognition test:** "Do you do this?"
|
|
142
|
+
- **Checklist application:** Rate experience with each pattern (radar chart)
|
|
143
|
+
- **Use in training:** Do novices improve after learning the pattern?
|
|
144
|
+
- **Refinement iteration:** Patterns are never "done", iterate based on feedback
|
|
145
|
+
|
|
146
|
+
**Quality signals:**
|
|
147
|
+
|
|
148
|
+
- Pattern feels obvious once articulated ("yes, exactly!")
|
|
149
|
+
- Practitioners can immediately give examples
|
|
150
|
+
- Novices can apply it and get results
|
|
151
|
+
- Pattern survives contact with reality (edge cases)
|
|
152
|
+
|
|
153
|
+
## Pattern Mining Techniques
|
|
154
|
+
|
|
155
|
+
### Interview Structure
|
|
156
|
+
|
|
157
|
+
**Pre-interview:**
|
|
158
|
+
|
|
159
|
+
- Define domain scope clearly
|
|
160
|
+
- Prepare concrete scenarios
|
|
161
|
+
- Schedule 60-90 minutes
|
|
162
|
+
- Record (with permission) for later analysis
|
|
163
|
+
|
|
164
|
+
**During interview:**
|
|
165
|
+
|
|
166
|
+
- Start with recent, vivid examples
|
|
167
|
+
- Follow energy (what they're excited about)
|
|
168
|
+
- Probe vague language ("elegant" → "what specifically?")
|
|
169
|
+
- Ask for counter-examples
|
|
170
|
+
|
|
171
|
+
**Post-interview:**
|
|
172
|
+
|
|
173
|
+
- Transcribe key quotes
|
|
174
|
+
- Note emotional peaks (often reveal values)
|
|
175
|
+
- Extract concrete actions, not philosophies
|
|
176
|
+
- Map to other interviews (look for patterns)
|
|
177
|
+
|
|
178
|
+
### Observation Protocol
|
|
179
|
+
|
|
180
|
+
If you can watch experts work:
|
|
181
|
+
|
|
182
|
+
1. **Shadow silently** - don't interrupt flow
|
|
183
|
+
2. **Note micro-behaviors** - what they check, order of operations
|
|
184
|
+
3. **Ask after** - "Why did you do X before Y?"
|
|
185
|
+
4. **Look for routines** - repeated sequences that might be unconscious
|
|
186
|
+
|
|
187
|
+
### Collaborative Pattern Writing
|
|
188
|
+
|
|
189
|
+
**Pattern: COLLABORATIVE CREATION**
|
|
190
|
+
Write patterns together with domain experts, not alone after interviews.
|
|
191
|
+
|
|
192
|
+
**Workshop format:**
|
|
193
|
+
|
|
194
|
+
- Gather 5-10 practitioners
|
|
195
|
+
- Share initial pattern drafts
|
|
196
|
+
- Discuss, refine, merge, split
|
|
197
|
+
- Experts challenge each other's assumptions
|
|
198
|
+
- Converge on shared understanding
|
|
199
|
+
|
|
200
|
+
**Benefits:**
|
|
201
|
+
|
|
202
|
+
- Experts trigger each other's memories
|
|
203
|
+
- Disagreements reveal context boundaries
|
|
204
|
+
- Shared ownership increases adoption
|
|
205
|
+
- Faster validation loop
|
|
206
|
+
|
|
207
|
+
## Writing Patterns
|
|
208
|
+
|
|
209
|
+
### Style Guidelines
|
|
210
|
+
|
|
211
|
+
**Use imperative mood:**
|
|
212
|
+
|
|
213
|
+
- Good: "Interview experienced people"
|
|
214
|
+
- Bad: "One should interview experienced people"
|
|
215
|
+
|
|
216
|
+
**Be specific, not abstract:**
|
|
217
|
+
|
|
218
|
+
- Good: "Select 5-10 people with different specialties"
|
|
219
|
+
- Bad: "Ensure diverse representation"
|
|
220
|
+
|
|
221
|
+
**Show, don't just tell:**
|
|
222
|
+
|
|
223
|
+
- Include examples, quotes, concrete cases
|
|
224
|
+
- "For instance, when creating the Learning Patterns..."
|
|
225
|
+
|
|
226
|
+
**Short paragraphs, scannable:**
|
|
227
|
+
|
|
228
|
+
- Patterns are reference material, not novels
|
|
229
|
+
- Use bullets, bold, subheadings
|
|
230
|
+
- Target 1-2 pages per pattern
|
|
231
|
+
|
|
232
|
+
**Evocative names:**
|
|
233
|
+
|
|
234
|
+
- Capitalized pattern names (PATTERN NAME)
|
|
235
|
+
- Create shared vocabulary
|
|
236
|
+
- Names should be memorable, slightly poetic
|
|
237
|
+
|
|
238
|
+
### Pattern Template
|
|
239
|
+
|
|
240
|
+
```
|
|
241
|
+
## PATTERN NAME
|
|
242
|
+
|
|
243
|
+
**Context:** When does this apply?
|
|
244
|
+
|
|
245
|
+
**Problem:** What tension exists?
|
|
246
|
+
|
|
247
|
+
**Forces:**
|
|
248
|
+
- Force A (e.g., need speed)
|
|
249
|
+
- Force B (e.g., need accuracy)
|
|
250
|
+
|
|
251
|
+
**Solution:**
|
|
252
|
+
Actionable description of the pattern.
|
|
253
|
+
Step-by-step if appropriate.
|
|
254
|
+
|
|
255
|
+
**Examples:**
|
|
256
|
+
Concrete instances of the pattern in practice.
|
|
257
|
+
|
|
258
|
+
**Consequences:**
|
|
259
|
+
- Benefits: What you gain
|
|
260
|
+
- Liabilities: What you trade off
|
|
261
|
+
|
|
262
|
+
**Related patterns:**
|
|
263
|
+
- Uses: X, Y
|
|
264
|
+
- Refines: Z
|
|
265
|
+
- See also: A, B
|
|
266
|
+
```
|
|
267
|
+
|
|
268
|
+
## Validation Strategies
|
|
269
|
+
|
|
270
|
+
### Recognition Test
|
|
271
|
+
|
|
272
|
+
Show pattern to 5+ practitioners who weren't interviewed:
|
|
273
|
+
|
|
274
|
+
- "Do you recognize this in your practice?"
|
|
275
|
+
- "Can you give me an example from your work?"
|
|
276
|
+
|
|
277
|
+
If < 60% recognize it, pattern might be too specific or poorly described.
|
|
278
|
+
|
|
279
|
+
### Application Test
|
|
280
|
+
|
|
281
|
+
Give pattern to novices:
|
|
282
|
+
|
|
283
|
+
- Can they apply it without expert guidance?
|
|
284
|
+
- Does performance improve measurably?
|
|
285
|
+
- What confusions arise? (reveals missing context)
|
|
286
|
+
|
|
287
|
+
### Radar Chart Review
|
|
288
|
+
|
|
289
|
+
**Pattern: PATTERN-EXPERIENCE CHART**
|
|
290
|
+
Create checklist of all patterns. Have practitioners rate their experience with each (0-5).
|
|
291
|
+
|
|
292
|
+
Plot as radar chart. Reveals:
|
|
293
|
+
|
|
294
|
+
- Which patterns are universal vs specialized
|
|
295
|
+
- Gaps in individual expertise (learning opportunities)
|
|
296
|
+
- Clusters of related patterns
|
|
297
|
+
|
|
298
|
+
### Iteration Signals
|
|
299
|
+
|
|
300
|
+
**Refine pattern if:**
|
|
301
|
+
|
|
302
|
+
- Multiple people misinterpret it
|
|
303
|
+
- Edge cases keep arising
|
|
304
|
+
- Name doesn't stick (people paraphrase it)
|
|
305
|
+
- Consequences list is lopsided (all benefits, no trade-offs)
|
|
306
|
+
|
|
307
|
+
**Split pattern if:**
|
|
308
|
+
|
|
309
|
+
- Two distinct solutions mixed together
|
|
310
|
+
- Context variations create different solutions
|
|
311
|
+
- Pattern is > 3 pages
|
|
312
|
+
|
|
313
|
+
**Merge patterns if:**
|
|
314
|
+
|
|
315
|
+
- Always used together
|
|
316
|
+
- Distinction feels artificial
|
|
317
|
+
- Practitioners can't tell them apart
|
|
318
|
+
|
|
319
|
+
## Meta-Patterns for Skill Creation
|
|
320
|
+
|
|
321
|
+
### Pattern: SKILL AS PATTERN LANGUAGE
|
|
322
|
+
|
|
323
|
+
When creating a skill, treat it as a mini pattern language:
|
|
324
|
+
|
|
325
|
+
- Each section is a pattern (context → problem → solution)
|
|
326
|
+
- Patterns connect and reference each other
|
|
327
|
+
- Skill header is the "overview" pattern
|
|
328
|
+
|
|
329
|
+
### Pattern: EXTRACT FROM HISTORY
|
|
330
|
+
|
|
331
|
+
When creating skills for yourself:
|
|
332
|
+
|
|
333
|
+
- Search CASS for past sessions in the domain
|
|
334
|
+
- Look for what worked vs what failed
|
|
335
|
+
- Extract patterns from your own tacit knowledge
|
|
336
|
+
- Your debugging sessions reveal troubleshooting patterns
|
|
337
|
+
|
|
338
|
+
### Pattern: REFERENCE OVER TUTORIAL
|
|
339
|
+
|
|
340
|
+
Skills are reference material, not step-by-step tutorials:
|
|
341
|
+
|
|
342
|
+
- Imperative, scannable, indexed
|
|
343
|
+
- Practitioners should find what they need in < 30 seconds
|
|
344
|
+
- Front-load the most valuable patterns
|
|
345
|
+
|
|
346
|
+
### Pattern: LIVING DOCUMENT
|
|
347
|
+
|
|
348
|
+
Patterns evolve as practice evolves:
|
|
349
|
+
|
|
350
|
+
- Add new patterns when gaps discovered
|
|
351
|
+
- Mark deprecated patterns when practice changes
|
|
352
|
+
- Track pattern maturity (candidate → established → proven → deprecated)
|
|
353
|
+
- Version skills, note major changes
|
|
354
|
+
|
|
355
|
+
## Quick Reference
|
|
356
|
+
|
|
357
|
+
**Creating a pattern language:**
|
|
358
|
+
|
|
359
|
+
1. Select 5-10 experienced, diverse practitioners
|
|
360
|
+
2. Interview for concrete experiences, not opinions
|
|
361
|
+
3. Mine recurring themes across interviews
|
|
362
|
+
4. Draft patterns (context, problem, solution, consequences)
|
|
363
|
+
5. Validate with practitioners via recognition/application tests
|
|
364
|
+
6. Iterate based on feedback
|
|
365
|
+
7. Document pattern relationships
|
|
366
|
+
|
|
367
|
+
**Pattern quality checklist:**
|
|
368
|
+
|
|
369
|
+
- [ ] Name is evocative and memorable
|
|
370
|
+
- [ ] Context clearly defines when to apply
|
|
371
|
+
- [ ] Problem describes the tension
|
|
372
|
+
- [ ] Solution is actionable and specific
|
|
373
|
+
- [ ] Consequences include both benefits and liabilities
|
|
374
|
+
- [ ] Examples are concrete, not hypothetical
|
|
375
|
+
- [ ] Pattern is 1-2 pages max
|
|
376
|
+
- [ ] 60%+ of practitioners recognize it
|
|
377
|
+
|
|
378
|
+
**Common mistakes:**
|
|
379
|
+
|
|
380
|
+
- Asking what people think instead of what they do
|
|
381
|
+
- Documenting aspirations instead of actual practice
|
|
382
|
+
- Writing abstractions instead of concrete actions
|
|
383
|
+
- Skipping validation (your draft is always wrong)
|
|
384
|
+
- Over-generalizing from one expert's approach
|
|
385
|
+
- Creating too many patterns (paralysis)
|
|
386
|
+
|
|
387
|
+
**Rule of thumb:** If experts say "well, duh" when reading the pattern, you got it right. If they say "I've never done that", you invented instead of discovered.
|