UNPKG

@alavida/agentpack

Version:

Compiler-driven lifecycle CLI for source-backed agent skills

docs.alavida.ai

alavida-ai/agentpack

118 lines (80 loc) • 2.75 kB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
--- name: identifying-skill-opportunities description: Use when deciding what raw knowledge should become a packaged skill, a reusable capability boundary, or an import edge in an agentpack skillgraph. type: core library: agentpack library_version: "0.1.10" sources: - "alavida-ai/agentpack:docs/how-it-works.mdx" - "alavida-ai/agentpack:docs/introduction.mdx" - "alavida-ai/agentpack:skills/agentpack-cli/SKILL.md" --- # Agentpack - Identifying Skill Opportunities ## Setup ```md Source knowledge: - domains/design/knowledge/brand-guidelines.md - domains/frontend/knowledge/component-heuristics.md Reusable skills: - domains/design/skills/brand-guidelines - domains/frontend/skills/frontend-skill Composed task skill: - domains/design/skills/agonda-brand-frontend imports: - @scope/brand-guidelines - @scope/frontend-skill ``` ## Core Patterns ### Split by reusable capability If a knowledge area should be reused compositionally in work, give it its own packaged skill. ### Compose task-specific skills with `import` Use a task skill when the work needs several reusable capabilities together. ### Keep capability boundaries explicit Use packaged skills and explicit `import` edges to expose reusable capabilities. Runtime materialization belongs downstream from the compiled graph. ## Common Mistakes ### HIGH Copying knowledge into one giant skill Wrong: ```md One SKILL.md contains branding, frontend heuristics, research notes, and delivery instructions. ``` Correct: ```md Create packaged skills for reusable capabilities, then compose them with imports. ``` Flattening multiple capabilities into one skill destroys explicit dependency edges and makes maintenance coarse. Source: maintainer interview ### CRITICAL Hiding dependency boundaries inside one task skill Wrong: ```text domains/brand/skills/website-ops/SKILL.md ``` Correct: ```text domains/brand/skills/copywriting/SKILL.md domains/research/skills/interview-research/SKILL.md domains/brand/skills/website-ops/SKILL.md # imports packaged skills ``` The compiled skill graph is the architectural source of truth. Reusable capability boundaries should remain explicit in packaged skills and `import` edges. Source: docs/architecture.mdx ### CRITICAL Omitting provenance sources for knowledge-backed skills Wrong: ```markdown --- name: value-copywriting description: Copy. --- ``` Correct: ```markdown --- name: value-copywriting description: Copy. --- ```agentpack source tone = "domains/value/knowledge/tone-of-voice.md" ``` ``` Without `source` bindings, the skillgraph cannot explain stale state against source truth. Source: docs/architecture.mdx ## References - [Capability boundaries](references/capability-boundaries.md)