mindlore 0.2.1 → 0.3.1

package/skills/mindlore-explore/SKILL.md

@@ -0,0 +1,104 @@
+---
+name: mindlore-explore
+description: Discover unexpected connections between sources — undirected knowledge exploration
+effort: medium
+allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Agent]
+---
+
+# /mindlore-explore
+
+Discover unexpected connections between knowledge sources. Undirected exploration — unlike query (directed search).
+
+## Scope
+
+Determine target using `getActiveMindloreDir()` logic:
+- Default (no flag): explore project `.mindlore/`
+- `--global`: explore `~/.mindlore/` + project cross-reference (most valuable mode)
+- Never hardcode `.mindlore/` path — always resolve dynamically
+
+## Trigger
+
+User says `/mindlore-explore`, "knowledge explore", "baglanti kesfet", "cross-reference bul".
+
+## Flow
+
+**Agent Delegation:** Cross-reference analizini subagent'a delege et (context koruma).
+
+1. Spawn an Agent for analysis:
+   ```
+   Agent({
+     description: "mindlore explore: connections",
+     subagent_type: "Explore",
+     prompt: "[mindlore:explore] Analyze .mindlore/ for unexpected connections. <aşağıdaki talimatları buraya koy>"
+   })
+   ```
+
+   Agent talimatları:
+   a. Read all source and domain files from active scope
+   b. Cross-match by tag + content (see criteria below)
+   c. Rank connections by strength
+   d. Return findings as structured table
+
+2. After agent returns — review and show findings to user
+3. On approval, write connection files (main session handles writes)
+
+**Cross-match criteria:**
+   - Files sharing tags but not referencing each other
+   - Sources covering similar topics from different angles
+   - Sources that could bridge between domains
+3. Rank connections by strength:
+   - **high**: 3+ shared tags + content overlap
+   - **medium**: 2 shared tags or significant content similarity
+   - **low**: 1 shared tag + weak content match
+4. Show findings to user
+5. On approval, write to `connections/` directory
+
+## Connection File Format
+
+Written to `connections/` with frontmatter:
+
+```markdown
+---
+type: connection
+slug: connection-source-a-source-b
+date_created: 2026-04-12
+sources: [source-a.md, source-b.md]
+domains: [domain-x.md]
+strength: high
+tags: [shared-tag-1, shared-tag-2]
+---
+
+## Connection
+
+[Why these sources are related — LLM explanation]
+
+## Action Suggestion
+
+[What could be done — domain update, new analysis, etc.]
+```
+
+## Rules
+
+- Check for duplicate connections before writing (same source pair)
+- Show findings before writing — user approval required
+- Update INDEX.md with new connections entry
+- Append EXPLORE entry to log.md
+- Strength is LLM-assessed based on tag overlap + content similarity
+- The `[mindlore:explore]` marker in the Agent prompt is required — it triggers the model-router hook to use the cost-optimized model (sonnet by default)
+
+## Output Format
+
+```
+Mindlore Explore — Discovered Connections
+
+Scope: project (.mindlore/)
+Sources scanned: 12
+Connections found: 3 (2 new, 1 existing)
+
+| # | Sources | Strength | Why |
+|---|---------|----------|-----|
+| 1 | react-hooks.md + agent-orchestration.md | medium | Both discuss state management patterns |
+| 2 | karpathy-kb.md + search-retrieval.md | high | FTS5 + knowledge architecture overlap |
+
+Write connections? (y/n)
+```

package/skills/mindlore-health/SKILL.md

@@ -9,6 +9,14 @@ allowed-tools: [Bash, Read]
 
 Run the 16-point structural health check on the `.mindlore/` knowledge base.
 
+## Scope
+
+Determine target using `getActiveMindloreDir()` logic:
+- Default (no flag): check project `.mindlore/`, fall back to global
+- `--global`: check only `~/.mindlore/`
+- `--all`: check both project + global scopes, report each separately
+- Never hardcode `.mindlore/` path — always resolve dynamically
+
 ## Trigger
 
 User says "health check", "mindlore health", "bilgi sistemi kontrol", "saglik kontrolu".

package/skills/mindlore-ingest/SKILL.md

@@ -9,6 +9,14 @@ allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Agent, WebFetch]
 
 Add a new knowledge source to the `.mindlore/` knowledge base.
 
+## Scope
+
+Determine target directory using `getActiveMindloreDir()` logic:
+- If CWD has `.mindlore/` → write to project scope
+- Otherwise → write to global `~/.mindlore/`
+- `--global` flag: force write to `~/.mindlore/` even if project scope exists
+- Never hardcode `.mindlore/` path — always resolve dynamically
+
 ## Trigger
 
 User shares a URL, text, file, or says "kaynak ekle", "source ingest", "bu linki kaydet", "knowledge ingest".
@@ -16,37 +24,59 @@ User shares a URL, text, file, or says "kaynak ekle", "source ingest", "bu linki
 ## Modes
 
 ### URL Mode
-1. Extract content from URL:
-   - If `markitdown` is available: `markitdown <url>` (best quality, zero tokens)
-   - Else: use `WebFetch` or `ctx_fetch_and_index`
-2. Save raw capture to `.mindlore/raw/` with frontmatter:
-   ```yaml
-   ---
-   slug: source-name-kebab
-   type: raw
-   source_url: https://...
-   date_captured: YYYY-MM-DD
-   tags: [tag1, tag2]
-   ---
+
+**Agent Delegation:** URL fetch + raw/sources yazımını subagent'a delege et (maliyet optimizasyonu).
+
+1. Spawn an Agent with the following pattern:
    ```
-3. Summarize into `.mindlore/sources/` with full frontmatter:
-   ```yaml
-   ---
-   slug: source-name-kebab
-   type: source
-   title: Human Readable Title
-   source_url: https://...
-   source_type: github-repo|blog|docs|video|x-thread
-   date_captured: YYYY-MM-DD
-   tags: [tag1, tag2]
-   quality: high|medium|low
-   ingested: true
-   ---
+   Agent({
+     description: "mindlore ingest: <slug>",
+     subagent_type: "researcher",
+     prompt: "[mindlore:ingest] <aşağıdaki talimatları buraya koy>"
+   })
    ```
-4. Update relevant domain page(s) in `.mindlore/domains/` (max 2)
-5. Update `.mindlore/INDEX.md` stats line
-6. Append entry to `.mindlore/log.md`
-7. Run FTS5 re-index: `npm run index`
+
+   Agent talimatları:
+   - Extract content from URL:
+     - If `markitdown` is available: `markitdown <url>` (best quality, zero tokens)
+     - Else: use `WebFetch` or `ctx_fetch_and_index`
+     - **YouTube URL** detected (`youtube.com` or `youtu.be`):
+       1. markitdown installed → `markitdown <url>` (includes transcript)
+       2. Else youtube-transcript npm → `hasYoutubeTranscript()` check
+       3. Else → return error, ask user to paste transcript
+   - Save raw capture to `.mindlore/raw/` with frontmatter:
+     ```yaml
+     ---
+     slug: source-name-kebab
+     type: raw
+     source_url: https://...
+     date_captured: YYYY-MM-DD
+     tags: [tag1, tag2]
+     ---
+     ```
+   - Summarize into `.mindlore/sources/` with full frontmatter:
+     ```yaml
+     ---
+     slug: source-name-kebab
+     type: source
+     title: Human Readable Title
+     source_url: https://...
+     source_type: github-repo|blog|docs|video|x-thread
+     date_captured: YYYY-MM-DD
+     tags: [tag1, tag2]
+     quality: high|medium|low
+     ingested: true
+     ---
+     ```
+   - Agent must report: created file paths, slug, quality assigned
+
+2. After agent returns — verify raw/ and sources/ files exist and have valid frontmatter
+3. Update relevant domain page(s) in `.mindlore/domains/` (max 2)
+4. Update `.mindlore/INDEX.md` stats line
+5. Append entry to `.mindlore/log.md`
+6. Run FTS5 re-index: `npm run index`
+
+**IMPORTANT:** The `[mindlore:ingest]` marker in the Agent prompt is required — it triggers the model-router hook to use the cost-optimized model (haiku by default).
 
 ### Text Mode
 1. User pastes text directly
@@ -75,9 +105,12 @@ The sources/ file should contain:
 
 ## Quality Assessment
 
-- `high`: Primary source, authoritative, detailed, directly relevant
-- `medium`: Useful but secondary, partial coverage, or tangentially relevant
-- `low`: Reference only, outdated, or low signal-to-noise
+Assign quality automatically during ingest using this heuristic:
+- `high`: Official docs (anthropic, github docs, MDN), primary research, authoritative references
+- `medium`: Blog posts, tutorials, conference talks, X threads with substance
+- `low`: Raw notes, text pastes, quick captures, low signal-to-noise
+
+LLM may override the heuristic based on content analysis. Always set quality — never leave it null.
 
 ## Domain Update Rules
 

package/skills/mindlore-log/SKILL.md

@@ -2,6 +2,14 @@
 
 Session logging, pattern extraction, and wiki updates.
 
+## Scope
+
+Determine target using `getActiveMindloreDir()` logic:
+- If CWD has `.mindlore/` → write to project scope
+- Otherwise → write to global `~/.mindlore/`
+- Reflect mode: scans both project + global diary/ for patterns
+- Never hardcode `.mindlore/` path — always resolve dynamically
+
 ## Trigger
 
 `/mindlore-log <mode>` where mode is `log`, `reflect`, `status`, or `save`.
@@ -30,19 +38,28 @@ date: 2026-04-11
 
 ### reflect
 
-Scan old deltas, extract patterns, write to learnings/.
-
-**Flow (v0.2 — basic):**
-1. Read all non-archived delta files in `diary/` (no `archived: true` frontmatter)
-2. Present summary to user: "Found N unprocessed deltas spanning DATE1 to DATE2"
-3. For each delta, extract: repeated topics, recurring decisions, common file changes
-4. Propose learnings to user: "I found these patterns: ..."
-5. User approves → write to `learnings/{topic}.md` (append if exists, create if not)
-6. Format: `YAPMA:` / `BEST PRACTICE:` / `KRITIK:` prefixed rules
-7. Mark processed deltas: add `archived: true` to their frontmatter
-8. Append to `log.md`: `| {date} | reflect | {N} deltas processed, {M} learnings written |`
-
-**Important:** Do NOT auto-extract patterns. Present findings, user approves. This is v0.2 basic mode — automated pattern extraction deferred to v0.2.1.
+LLM-driven pattern extraction from diary deltas → persistent learnings.
+
+**Flow (v0.3 — LLM-driven):**
+1. Read last N non-archived delta files in `diary/` (no `archived: true` frontmatter)
+2. Present summary: "Found N unprocessed deltas spanning DATE1 to DATE2"
+3. LLM analyzes deltas for repeating patterns:
+   - Recurring topics/themes across sessions
+   - Repeated decisions (same choice made multiple times)
+   - Common mistakes or friction points
+   - Workflow patterns that worked well
+4. Categorize found patterns by topic (git, testing, architecture, etc.)
+5. Present findings to user with proposed learnings
+6. User approves → write to `learnings/{topic}.md` (append if exists, create if not)
+7. Format: `YAPMA:` / `BEST PRACTICE:` / `KRITIK:` prefixed rules
+8. Update relevant domain page if pattern relates to an existing domain
+9. Mark processed deltas: add `archived: true` to their frontmatter
+10. Append to `log.md`: `| {date} | reflect | {N} deltas processed, {M} learnings written |`
+
+**Rules:**
+- NEVER write learnings without user approval
+- Group related patterns into existing topic files (don't create one file per pattern)
+- Reflect scans both project + global diary/ in `--all` mode
 
 ### status
 

package/skills/mindlore-query/SKILL.md

@@ -2,6 +2,14 @@
 
 Search, ask, analyze, and retrieve knowledge from `.mindlore/`.
 
+## Scope
+
+Determine search scope using `getActiveMindloreDir()` / `getAllDbs()` logic:
+- Default (no flag): search project `.mindlore/` DB, fall back to global
+- `--global`: search only `~/.mindlore/` DB
+- `--all`: search both project + global DBs, project results first
+- Never hardcode `.mindlore/` path — always resolve dynamically
+
 ## Trigger
 
 `/mindlore-query <mode> [query]` where mode is `search`, `ask`, `stats`, or `brief`.

package/templates/SCHEMA.md

@@ -28,6 +28,18 @@ with YAML frontmatter. Search is powered by FTS5 (SQLite full-text search).
 └── mindlore.db     # FTS5 search database (SQLite)
 ```
 
+### Global Scope (v0.3)
+
+Mindlore supports two scopes:
+- **Project scope:** `.mindlore/` in the current working directory (project-specific knowledge)
+- **Global scope:** `~/.mindlore/` in the user's home directory (cross-project knowledge)
+
+Rules:
+- `getActiveMindloreDir()` resolves scope: project if exists, otherwise global
+- Search is layered: project results first, global results second
+- Skills accept `--global` (force global) and `--all` (both scopes) flags
+- `npx mindlore init --global` creates `~/.mindlore/` with git repo for auto-sync
+
 ### Directory Rules
 
 - Each directory corresponds to exactly one frontmatter `type` value
@@ -143,7 +155,7 @@ Discover unexpected connections between sources. Cross-reference analysis.
 - Max results: 3 per query (BM25 ranking)
 - Hook injects: file path + first 2 headings
 
-### FTS5 Columns (9-col schema, v0.2)
+### FTS5 Columns (10-col schema, v0.3)
 
 | Column | Indexed | Source |
 |--------|---------|--------|
@@ -155,7 +167,8 @@ Discover unexpected connections between sources. Cross-reference analysis.
 | `title` | Yes | Frontmatter title or first heading |
 | `content` | Yes | Markdown body (sans frontmatter) |
 | `tags` | Yes | Frontmatter tags (comma-separated) |
-| `quality` | UNINDEXED | Frontmatter quality (NULL until 50+ sources) |
+| `quality` | UNINDEXED | Frontmatter quality (high/medium/low) |
+| `date_captured` | UNINDEXED | Frontmatter date_captured or date |
 
 ### Search Flow (UserPromptSubmit hook)
 

package/templates/config.json

@@ -0,0 +1,9 @@
+{
+  "version": "0.3.1",
+  "models": {
+    "ingest": "haiku",
+    "evolve": "sonnet",
+    "explore": "sonnet",
+    "default": "haiku"
+  }
+}