persona-knowledge

Name: persona-knowledge
Author: acnlabs

40stars5forks

How to install persona-knowledge?

acnlabs/openpersona/persona-knowledge

$npx -y skills add acnlabs/OpenPersona --skill persona-knowledge

Installs into the current project.

Summary

Persistent, incremental, searchable persona knowledge base. Ingests data from Obsidian vaults, chat exports, X/Twitter archives, and more into a MemPalace-backed store with a Karpathy LLM Wiki knowledge layer. Exports training/ directories for persona-model-trainer.

Files · 1

View on GitHub

SKILL.md

1	# persona-knowledge
2
3	Persistent, incremental, searchable persona knowledge base — the data layer between raw sources and persona training.
4
5	Architecture: MemPalace (storage + search) + Knowledge Graph (relationships + timeline) + Karpathy LLM Wiki (knowledge accumulation)
6
7	Dependency chain: `data sources` → `persona-knowledge` → `anyone-skill` / `persona-model-trainer`
8
9	---
10
11	## When to use this skill
12
13	Trigger phrases:
14
15	- "create a dataset for this persona"
16	- "add data to the dataset"
17	- "import my Obsidian vault"
18	- "import my Twitter archive"
19	- "build a knowledge base for X"
20	- "export training data"
21	- "search the persona dataset"
22
23	Not suitable when:
24
25	- User wants a quick one-shot distillation without persistent storage (use anyone-skill alone)
26	- User only has < 50 messages of data (too little to warrant a dataset)
27
28	---
29
30	## Phase 1: Init
31
32	Create a new persona dataset:
33
34	```bash
35	python scripts/init_knowledge.py --slug {slug} --name "Display Name"
36	```
37
38	This creates `~/.openpersona/knowledge/{slug}/` with:
39
40	```
41	~/.openpersona/knowledge/{slug}/
42	dataset.json # metadata: slug, name, created_at, stats
43	.mempalace/ # MemPalace local data (per-dataset isolation via palace_path)
44	palace/ # MemPalace internal store (ChromaDB + KG)
45	sources/ # immutable source file backups
46	.source-index.json # per-file metadata: hash, import time, line count, PII flags
47	wiki/ # Karpathy wiki (LLM-maintained derived artifact)
48	_schema.md # wiki maintenance rules
49	identity.md
50	voice.md
51	values.md
52	thinking.md
53	relationships.md # generated from Knowledge Graph
54	timeline.md # generated from Knowledge Graph
55	_contradictions.md
56	_changelog.md
57	_evidence.md
58	```
59
60	MemPalace palace structure:
61	- One Wing per persona (named by slug)
62	- Halls mapped to 5 persona dimensions:
63	- `hall_facts` — Identity (background, career, education)
64	- `hall_events` — Memory (key events, turning points)
65	- `hall_preferences` — Personality (values, preferences, boundaries)
66	- `hall_discoveries` — Procedure (mental models, decision heuristics)
67	- `hall_voice` — Interaction (vocabulary, rhythm, humor, emotional temperature)
68
69	Gate: Confirm slug and display name with the user before proceeding.
70
71	---
72
73	## Phase 2: Ingest
74
75	Import data sources into the dataset. Can be called multiple times for incremental ingestion.
76
77	```bash
78	python scripts/ingest.py --slug {slug} --source <path> [--adapter <name>] [--since <date>]
79	```
80
81	Three adapters cover all supported formats:
82
83	\| Source \| Adapter \| Detection \|
84	\|--------\|---------\|-----------\|
85	\| Obsidian vault \| `universal` \| Directory containing `.obsidian/` or `*.md` files \|
86	\| GBrain export dir \| `universal` \| Markdown directory with `.raw/` sidecar dirs \|
87	\| `.md` / `.txt` / `.csv` / `.pdf` \| `universal` \| File extension \|
88	\| `.jsonl` / `.json` \| `universal` \| File extension \|
89	\| GBrain JSON export \| `universal` \| `.json` with `memories` key or `--entity` flag \|
90	\| WhatsApp `.txt` export \| `chat_export` \| Matches WhatsApp timestamp pattern \|
91	\| Telegram `result.json` \| `chat_export` \| JSON with `chats` key \|
92	\| Signal export \| `chat_export` \| JSON with Signal message format \|
93	\| iMessage `.db` \| `chat_export` \| SQLite with `message` + `handle` tables \|
94	\| X (Twitter) archive \| `social` \| Directory containing `data/tweets.js` \|
95	\| Instagram archive \| `social` \| Directory containing `content/posts_1.json` \|
96
97
98	Ingest pipeline (per source):
99
100	1. Parse — adapter converts source to unified `[{role, content, timestamp, source_file, source_type}]`
101	2. PII scan — flag SSN, credit card, email, password patterns
102	3. Hash dedup — SHA-256 content hash, skip already-ingested entries
103	4. Write sources/ — save parsed data as JSONL backup (immutable, one file per source)
104	5. Store in MemPalace — verbatim text into ChromaDB via palace wing/hall structure
105	6. Extract KG triples — detect entities and relationships, write to Knowledge Graph with temporal validity
106	7. Report — print source name, message count, assistant turns, PII flags, new KG entities
107
108	After each source is ingested, report:
109
110	```
111	✅ whatsapp-2024.txt → 1,247 messages (892 assista

People who liked this also viewed

6 picks

Type

persona-knowledge

40stars5forks

View on GitHub

How to install persona-knowledge?

acnlabs/openpersona/persona-knowledge

$npx -y skills add acnlabs/OpenPersona --skill persona-knowledge

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# persona-knowledge
2
3	Persistent, incremental, searchable persona knowledge base — the data layer between raw sources and persona training.
4
5	Architecture: MemPalace (storage + search) + Knowledge Graph (relationships + timeline) + Karpathy LLM Wiki (knowledge accumulation)
6
7	Dependency chain: `data sources` → `persona-knowledge` → `anyone-skill` / `persona-model-trainer`
8
9	---
10
11	## When to use this skill
12
13	Trigger phrases:
14
15	- "create a dataset for this persona"
16	- "add data to the dataset"
17	- "import my Obsidian vault"
18	- "import my Twitter archive"
19	- "build a knowledge base for X"
20	- "export training data"
21	- "search the persona dataset"
22
23	Not suitable when:
24
25	- User wants a quick one-shot distillation without persistent storage (use anyone-skill alone)
26	- User only has < 50 messages of data (too little to warrant a dataset)
27
28	---
29
30	## Phase 1: Init
31
32	Create a new persona dataset:
33
34	```bash
35	python scripts/init_knowledge.py --slug {slug} --name "Display Name"
36	```
37
38	This creates `~/.openpersona/knowledge/{slug}/` with:
39
40	```
41	~/.openpersona/knowledge/{slug}/
42	dataset.json # metadata: slug, name, created_at, stats
43	.mempalace/ # MemPalace local data (per-dataset isolation via palace_path)
44	palace/ # MemPalace internal store (ChromaDB + KG)
45	sources/ # immutable source file backups
46	.source-index.json # per-file metadata: hash, import time, line count, PII flags
47	wiki/ # Karpathy wiki (LLM-maintained derived artifact)
48	_schema.md # wiki maintenance rules
49	identity.md
50	voice.md
51	values.md
52	thinking.md
53	relationships.md # generated from Knowledge Graph
54	timeline.md # generated from Knowledge Graph
55	_contradictions.md
56	_changelog.md
57	_evidence.md
58	```
59
60	MemPalace palace structure:
61	- One Wing per persona (named by slug)
62	- Halls mapped to 5 persona dimensions:
63	- `hall_facts` — Identity (background, career, education)
64	- `hall_events` — Memory (key events, turning points)
65	- `hall_preferences` — Personality (values, preferences, boundaries)
66	- `hall_discoveries` — Procedure (mental models, decision heuristics)
67	- `hall_voice` — Interaction (vocabulary, rhythm, humor, emotional temperature)
68
69	Gate: Confirm slug and display name with the user before proceeding.
70
71	---
72
73	## Phase 2: Ingest
74
75	Import data sources into the dataset. Can be called multiple times for incremental ingestion.
76
77	```bash
78	python scripts/ingest.py --slug {slug} --source <path> [--adapter <name>] [--since <date>]
79	```
80
81	Three adapters cover all supported formats:
82
83	\| Source \| Adapter \| Detection \|
84	\|--------\|---------\|-----------\|
85	\| Obsidian vault \| `universal` \| Directory containing `.obsidian/` or `*.md` files \|
86	\| GBrain export dir \| `universal` \| Markdown directory with `.raw/` sidecar dirs \|
87	\| `.md` / `.txt` / `.csv` / `.pdf` \| `universal` \| File extension \|
88	\| `.jsonl` / `.json` \| `universal` \| File extension \|
89	\| GBrain JSON export \| `universal` \| `.json` with `memories` key or `--entity` flag \|
90	\| WhatsApp `.txt` export \| `chat_export` \| Matches WhatsApp timestamp pattern \|
91	\| Telegram `result.json` \| `chat_export` \| JSON with `chats` key \|
92	\| Signal export \| `chat_export` \| JSON with Signal message format \|
93	\| iMessage `.db` \| `chat_export` \| SQLite with `message` + `handle` tables \|
94	\| X (Twitter) archive \| `social` \| Directory containing `data/tweets.js` \|
95	\| Instagram archive \| `social` \| Directory containing `content/posts_1.json` \|
96
97
98	Ingest pipeline (per source):
99
100	1. Parse — adapter converts source to unified `[{role, content, timestamp, source_file, source_type}]`
101	2. PII scan — flag SSN, credit card, email, password patterns
102	3. Hash dedup — SHA-256 content hash, skip already-ingested entries
103	4. Write sources/ — save parsed data as JSONL backup (immutable, one file per source)
104	5. Store in MemPalace — verbatim text into ChromaDB via palace wing/hall structure
105	6. Extract KG triples — detect entities and relationships, write to Knowledge Graph with temporal validity
106	7. Report — print source name, message count, assistant turns, PII flags, new KG entities
107
108	After each source is ingested, report:
109
110	```
111	✅ whatsapp-2024.txt → 1,247 messages (892 assista

People who liked this also viewed

6 picks

Type