recall

Name: recall
Author: tuan3w

32stars2forks

How to install recall?

tuan3w/obsidian-vault-agent/recall

$npx -y skills add tuan3w/obsidian-vault-agent --skill recall

Installs into the current project.

Summary

Spaced repetition and retrieval practice engine for the vault. Use when the user wants a review session, recall practice, to test what they remember, or to resurface notes. Triggers on "review session", "what should I review", "recall practice", "resurface notes", "spaced repetit

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Run an interactive spaced repetition session: build a mixed queue of retrieval
3	practice cards, inbox triage items, and a surprise cross-domain connection.
4	For each card, show only the title first — forcing recall before revealing
5	content (retrieval practice). Interleave domains to exploit the spacing and
6	interleaving effects. Surface a non-obvious connection the user hasn't made yet.
7	</Purpose>
8
9	<Use_When>
10	- User wants a review session ("what should I review?", "quiz me", "recall practice")
11	- User wants to resurface notes they haven't seen in a while
12	- User wants to triage their inbox
13	- User wants to discover connections across domains
14	- User asks "what do I remember about X?"
15	- Periodic knowledge maintenance (daily, weekly reviews)
16	</Use_When>
17
18	<Do_Not_Use_When>
19	- User wants to deeply process a specific note (use /process)
20	- User wants vault structure analysis (use /vault-graph)
21	- User wants to synthesize across a cluster (use /synthesize)
22	- User is actively reading or adding a new source note
23	</Do_Not_Use_When>
24
25	<Execution_Policy>
26	- INTERACTIVE — never skip ahead. Each card requires the user to respond before
27	the answer is shown. The friction IS the learning.
28	- Never show note content before the user attempts recall
29	- Prefer notes with updated_date older than 14 days (spacing effect)
30	- MUST interleave domains — draw from at least 3 different topic folders
31	- Use MCP tools (search_notes, get_frontmatter) when available; fall back to Grep/Glob
32	- Session size: default 5 retrieval cards, 2 inbox items, 1 surprise connection
33	(adjust to user's argument if provided)
34	- After each reveal, offer to update the note if it's stale
35	</Execution_Policy>
36
37	<Steps>
38
39	## Stage 1: BUILD THE QUEUE
40
41	### 1a. Retrieval Practice Cards (default 3–5)
42
43	Candidate selection using MCP:
44	```
45	search_notes(query="processing_status: processed OR processing_status: evergreen", limit=50)
46	search_notes(query="type: term OR type: note", limit=50)
47	```
48
49	Or fall back to Grep:
50	```
51	Grep(pattern="processing_status: (processed\|evergreen)", path="notes/", glob="*.md")
52	Grep(pattern="^type: (term\|note)$", path="notes/", glob="*.md")
53	```
54
55	Filter and rank candidates:
56	1. Prefer notes with `updated_date` older than 14 days — compute from today (2026-03-20)
57	2. Prefer notes with the `[ ](#anki-card)` anchor (marked for spaced repetition)
58	3. Prefer notes with `processing_status: processed` over `evergreen`
59	(evergreen notes are maintained; processed ones need re-testing)
60	4. Enforce domain interleaving: collect the topic folder for each candidate
61	(`notes/ml/`, `notes/psychology/`, `notes/startup/`, etc.) and ensure the
62	final selection spans at least 3 different folders
63	5. If fewer than 3 domains are available, accept 2 minimum
64
65	Read the frontmatter to confirm `updated_date` and `type`:
66	```
67	get_frontmatter(path="notes/PATH/filename.md")
68	```
69
70	### 1b. Inbox Triage Items (1–2)
71
72	Find the oldest unprocessed source notes:
73	```
74	search_notes(query="processing_status: inbox", limit=20)
75	```
76	Or:
77	```
78	Grep(pattern="processing_status: inbox", path="notes/", glob="*.md")
79	```
80
81	Filter to types: `paper`, `post`, `book`, `lecture`, `course`.
82	Sort by `created_date` ascending — oldest first.
83	Pick 1–2.
84
85	### 1c. Surprise Connection (1)
86
87	Goal: find two notes from DIFFERENT domain folders that share a concept keyword
88	but have NO wikilink between them.
89
90	Algorithm:
91	1. Pick a retrieval card already in the queue (it has topic keywords in its body)
92	2. Extract 2–3 key concept words from that note's title or content
93	3. Search a DIFFERENT domain folder for notes containing those keywords:
94	```
95	Grep(pattern="KEYWORD", path="notes/OTHER_DOMAIN/", glob="*.md")
96	```
97	4. Check that the two notes do NOT link to each other:
98	```
99	Grep(pattern="\\[\\[.TITLE.\\]\\]", path="notes/SOURCE_NOTE.md")
100	```
101	5. If no match, try a second keyword or a second candidate note
102	6. Present both notes with the shared concept as the "bridge"
103
104	Non-obvious pairings to prioritize (cross-domain > within-domain):
105	- ml/ ↔ psychology/ (learning algorithms ↔ cognitive science)
106	- startup/ ↔ design/ (product strategy ↔ UX)
107	- finance/ ↔ ml/ (optimization ↔ market dynamics)
108	- psychology/ ↔ startup/ (behavioral economics ↔ business)
109	- computer science/ ↔ logic/ (formal systems ↔ reasoning)
110
111	## Stage 2: PRESENT QUEUE OVERVIEW
112
113	Before

People who liked this also viewed

6 picks

Type

recall

32stars2forks

View on GitHub

How to install recall?

tuan3w/obsidian-vault-agent/recall

$npx -y skills add tuan3w/obsidian-vault-agent --skill recall

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	<Purpose>
2	Run an interactive spaced repetition session: build a mixed queue of retrieval
3	practice cards, inbox triage items, and a surprise cross-domain connection.
4	For each card, show only the title first — forcing recall before revealing
5	content (retrieval practice). Interleave domains to exploit the spacing and
6	interleaving effects. Surface a non-obvious connection the user hasn't made yet.
7	</Purpose>
8
9	<Use_When>
10	- User wants a review session ("what should I review?", "quiz me", "recall practice")
11	- User wants to resurface notes they haven't seen in a while
12	- User wants to triage their inbox
13	- User wants to discover connections across domains
14	- User asks "what do I remember about X?"
15	- Periodic knowledge maintenance (daily, weekly reviews)
16	</Use_When>
17
18	<Do_Not_Use_When>
19	- User wants to deeply process a specific note (use /process)
20	- User wants vault structure analysis (use /vault-graph)
21	- User wants to synthesize across a cluster (use /synthesize)
22	- User is actively reading or adding a new source note
23	</Do_Not_Use_When>
24
25	<Execution_Policy>
26	- INTERACTIVE — never skip ahead. Each card requires the user to respond before
27	the answer is shown. The friction IS the learning.
28	- Never show note content before the user attempts recall
29	- Prefer notes with updated_date older than 14 days (spacing effect)
30	- MUST interleave domains — draw from at least 3 different topic folders
31	- Use MCP tools (search_notes, get_frontmatter) when available; fall back to Grep/Glob
32	- Session size: default 5 retrieval cards, 2 inbox items, 1 surprise connection
33	(adjust to user's argument if provided)
34	- After each reveal, offer to update the note if it's stale
35	</Execution_Policy>
36
37	<Steps>
38
39	## Stage 1: BUILD THE QUEUE
40
41	### 1a. Retrieval Practice Cards (default 3–5)
42
43	Candidate selection using MCP:
44	```
45	search_notes(query="processing_status: processed OR processing_status: evergreen", limit=50)
46	search_notes(query="type: term OR type: note", limit=50)
47	```
48
49	Or fall back to Grep:
50	```
51	Grep(pattern="processing_status: (processed\|evergreen)", path="notes/", glob="*.md")
52	Grep(pattern="^type: (term\|note)$", path="notes/", glob="*.md")
53	```
54
55	Filter and rank candidates:
56	1. Prefer notes with `updated_date` older than 14 days — compute from today (2026-03-20)
57	2. Prefer notes with the `[ ](#anki-card)` anchor (marked for spaced repetition)
58	3. Prefer notes with `processing_status: processed` over `evergreen`
59	(evergreen notes are maintained; processed ones need re-testing)
60	4. Enforce domain interleaving: collect the topic folder for each candidate
61	(`notes/ml/`, `notes/psychology/`, `notes/startup/`, etc.) and ensure the
62	final selection spans at least 3 different folders
63	5. If fewer than 3 domains are available, accept 2 minimum
64
65	Read the frontmatter to confirm `updated_date` and `type`:
66	```
67	get_frontmatter(path="notes/PATH/filename.md")
68	```
69
70	### 1b. Inbox Triage Items (1–2)
71
72	Find the oldest unprocessed source notes:
73	```
74	search_notes(query="processing_status: inbox", limit=20)
75	```
76	Or:
77	```
78	Grep(pattern="processing_status: inbox", path="notes/", glob="*.md")
79	```
80
81	Filter to types: `paper`, `post`, `book`, `lecture`, `course`.
82	Sort by `created_date` ascending — oldest first.
83	Pick 1–2.
84
85	### 1c. Surprise Connection (1)
86
87	Goal: find two notes from DIFFERENT domain folders that share a concept keyword
88	but have NO wikilink between them.
89
90	Algorithm:
91	1. Pick a retrieval card already in the queue (it has topic keywords in its body)
92	2. Extract 2–3 key concept words from that note's title or content
93	3. Search a DIFFERENT domain folder for notes containing those keywords:
94	```
95	Grep(pattern="KEYWORD", path="notes/OTHER_DOMAIN/", glob="*.md")
96	```
97	4. Check that the two notes do NOT link to each other:
98	```
99	Grep(pattern="\\[\\[.TITLE.\\]\\]", path="notes/SOURCE_NOTE.md")
100	```
101	5. If no match, try a second keyword or a second candidate note
102	6. Present both notes with the shared concept as the "bridge"
103
104	Non-obvious pairings to prioritize (cross-domain > within-domain):
105	- ml/ ↔ psychology/ (learning algorithms ↔ cognitive science)
106	- startup/ ↔ design/ (product strategy ↔ UX)
107	- finance/ ↔ ml/ (optimization ↔ market dynamics)
108	- psychology/ ↔ startup/ (behavioral economics ↔ business)
109	- computer science/ ↔ logic/ (formal systems ↔ reasoning)
110
111	## Stage 2: PRESENT QUEUE OVERVIEW
112
113	Before

People who liked this also viewed

6 picks

Type