04-preference-and-memory

Name: 04-preference-and-memory
Author: heymegabyte

18stars4forks

How to install 04-preference-and-memory?

heymegabyte/claude-skills/04-preference-and-memory

$npx -y skills add heymegabyte/claude-skills --skill 04-preference-and-memory

Installs into the current project.

Summary

Captures and evolves user preferences with confidence levels. Maintains Voice of the Customer model with exact language, dissatisfaction and aspiration signals. Handles promotion/demotion, global vs project scoping, auto memory system, and Omi wearable data integration.

Files · 1

View on GitHub

SKILL.md

1	# 04 — Preference and Memory
2
3	Capture and evolve user preferences with confidence levels, maintaining a Voice of the Customer model scoped globally or per project.
4
5	## Auto-memory system
6
7	Persistent file-based memory at `/Users/Apple/.claude/projects/<encoded-path>/memory/`. Write directly via Write tool.
8
9	## Memory types
10
11	- user — role, goals, responsibilities, knowledge
12	- feedback — guidance on how to approach work (corrections AND confirmations)
13	- project — ongoing initiatives, bugs, incidents, who/why/when
14	- reference — pointers to external resources
15
16	## Capture triggers
17
18	- User explicitly says "remember X" / "save Y" → save immediately
19	- Correction ("no not that", "don't") → feedback memory
20	- Confirmation of unusual approach ("yes exactly", "perfect") → feedback memory
21	- New role / preference / responsibility detail → user memory
22	- Project milestone / decision / deadline → project memory
23	- External system reference ("check the Linear project X") → reference memory
24
25	## Memory file format
26
27	```markdown
28	---
29	name: {short-kebab-case-slug}
30	description: {one-line summary — used for relevance ranking}
31	metadata:
32	type: {user \| feedback \| project \| reference}
33	---
34
35	{memory content}
36	```
37
38	For feedback/project: lead with rule/fact, then `Why:` line + `How to apply:` line.
39
40	## Memory routing
41
42	- Universal (across all projects) → `~/.claude/`
43	- Project-specific → `./.claude/` (path-scoped)
44	- Project memory file → `~/.claude/projects/<encoded-path>/memory/<name>.md` + index in `MEMORY.md`
45
46	## MEMORY.md index
47
48	One-line pointer per file:
49
50	```md
51	- [Title](file.md) — one-line hook
52	```
53
54	Keep under 200 lines (lines after 200 truncated).
55
56	## Confidence levels
57
58	- 0.95-1.0 — directly stated by Brian
59	- 0.85-0.95 — strongly implied by pattern (3+ confirmations)
60	- 0.70-0.85 — inferred from one signal
61	- <0.70 — speculative, verify before acting
62
63	Below 0.70 → ask `AskUserQuestion` before persisting.
64
65	## Voice of the Customer model
66
67	- Exact language Brian uses ("pick ONE", "Hey not Hi", "always find 50 more things")
68	- Dissatisfaction signals (corrections, re-issued prompts)
69	- Aspiration signals (stated goals, ideal-world descriptions)
70	- Stylistic preferences (brevity, sharpness, no preamble)
71
72	## Brian's preferences (cite from `rules/brian-preferences.md`)
73
74	- Pick ONE, never options
75	- Never ask permission, silence = approval
76	- "Hey" not "Hi", no preamble
77	- Code complete, never truncate
78	- Priority: simplicity > cost > speed > compatibility
79	- Open-source only
80	- Side repos auto-push; emdash projects user-push
81
82	## Promotion / demotion
83
84	- Memory recurring across 3+ projects → promote to rule (`~/.claude/plugins/heymegabyte-claude-skills/rules/`)
85	- Rule contradicted by repeated correction → demote to memory or remove
86	- Per `rules/prompt-as-training-signal.md`
87
88	## Stale memory
89
90	- Memory describing now-incorrect facts (renamed function, removed flag, retired service) → update OR delete
91	- Per `rules/prompt-as-training-signal.md` § Before recommending from memory
92
93	## Omi wearable integration (future)
94
95	- Capture Brian's voice notes as candidate memory entries
96	- Auto-tag by inferred type
97	- Surface for review before commit
98	- Path: `~/.claude/_omi-inbox/<timestamp>.md`
99
100	## Capture protocol (every prompt)
101
102	1. Scan prompt for capture triggers
103	2. Determine memory type
104	3. Write file w/ frontmatter
105	4. Add index entry to `MEMORY.md`
106	5. Cross-link siblings via `[[name]]`
107	6. Per `rules/prompt-as-training-signal.md` — extract BEFORE doing requested work

People who liked this also viewed

6 picks

Type

04-preference-and-memory

18stars4forks

View on GitHub

How to install 04-preference-and-memory?

heymegabyte/claude-skills/04-preference-and-memory

$npx -y skills add heymegabyte/claude-skills --skill 04-preference-and-memory

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# 04 — Preference and Memory
2
3	Capture and evolve user preferences with confidence levels, maintaining a Voice of the Customer model scoped globally or per project.
4
5	## Auto-memory system
6
7	Persistent file-based memory at `/Users/Apple/.claude/projects/<encoded-path>/memory/`. Write directly via Write tool.
8
9	## Memory types
10
11	- user — role, goals, responsibilities, knowledge
12	- feedback — guidance on how to approach work (corrections AND confirmations)
13	- project — ongoing initiatives, bugs, incidents, who/why/when
14	- reference — pointers to external resources
15
16	## Capture triggers
17
18	- User explicitly says "remember X" / "save Y" → save immediately
19	- Correction ("no not that", "don't") → feedback memory
20	- Confirmation of unusual approach ("yes exactly", "perfect") → feedback memory
21	- New role / preference / responsibility detail → user memory
22	- Project milestone / decision / deadline → project memory
23	- External system reference ("check the Linear project X") → reference memory
24
25	## Memory file format
26
27	```markdown
28	---
29	name: {short-kebab-case-slug}
30	description: {one-line summary — used for relevance ranking}
31	metadata:
32	type: {user \| feedback \| project \| reference}
33	---
34
35	{memory content}
36	```
37
38	For feedback/project: lead with rule/fact, then `Why:` line + `How to apply:` line.
39
40	## Memory routing
41
42	- Universal (across all projects) → `~/.claude/`
43	- Project-specific → `./.claude/` (path-scoped)
44	- Project memory file → `~/.claude/projects/<encoded-path>/memory/<name>.md` + index in `MEMORY.md`
45
46	## MEMORY.md index
47
48	One-line pointer per file:
49
50	```md
51	- [Title](file.md) — one-line hook
52	```
53
54	Keep under 200 lines (lines after 200 truncated).
55
56	## Confidence levels
57
58	- 0.95-1.0 — directly stated by Brian
59	- 0.85-0.95 — strongly implied by pattern (3+ confirmations)
60	- 0.70-0.85 — inferred from one signal
61	- <0.70 — speculative, verify before acting
62
63	Below 0.70 → ask `AskUserQuestion` before persisting.
64
65	## Voice of the Customer model
66
67	- Exact language Brian uses ("pick ONE", "Hey not Hi", "always find 50 more things")
68	- Dissatisfaction signals (corrections, re-issued prompts)
69	- Aspiration signals (stated goals, ideal-world descriptions)
70	- Stylistic preferences (brevity, sharpness, no preamble)
71
72	## Brian's preferences (cite from `rules/brian-preferences.md`)
73
74	- Pick ONE, never options
75	- Never ask permission, silence = approval
76	- "Hey" not "Hi", no preamble
77	- Code complete, never truncate
78	- Priority: simplicity > cost > speed > compatibility
79	- Open-source only
80	- Side repos auto-push; emdash projects user-push
81
82	## Promotion / demotion
83
84	- Memory recurring across 3+ projects → promote to rule (`~/.claude/plugins/heymegabyte-claude-skills/rules/`)
85	- Rule contradicted by repeated correction → demote to memory or remove
86	- Per `rules/prompt-as-training-signal.md`
87
88	## Stale memory
89
90	- Memory describing now-incorrect facts (renamed function, removed flag, retired service) → update OR delete
91	- Per `rules/prompt-as-training-signal.md` § Before recommending from memory
92
93	## Omi wearable integration (future)
94
95	- Capture Brian's voice notes as candidate memory entries
96	- Auto-tag by inferred type
97	- Surface for review before commit
98	- Path: `~/.claude/_omi-inbox/<timestamp>.md`
99
100	## Capture protocol (every prompt)
101
102	1. Scan prompt for capture triggers
103	2. Determine memory type
104	3. Write file w/ frontmatter
105	4. Add index entry to `MEMORY.md`
106	5. Cross-link siblings via `[[name]]`
107	6. Per `rules/prompt-as-training-signal.md` — extract BEFORE doing requested work

People who liked this also viewed

6 picks

Type