@redaksjon/protokoll 0.1.0 → 0.3.0

package/guide/development.md

@@ -36,6 +36,15 @@ npm run lint
 npm run lint:fix  # Auto-fix issues
 ```
 
+## Recent Fixes
+
+### Tag Deduplication (2026-01-19)
+
+Tags in transcript metadata are now automatically deduplicated. Previously, when the same entity (e.g., "xenocline") was identified through multiple classification signals (e.g., both as an explicit phrase and as an associated project), it would appear multiple times in the tags array. Now tags are deduplicated using Set, ensuring each tag appears only once regardless of how many signals identify it.
+
+**Changed**: `src/util/metadata.ts` - `extractTagsFromSignals()` function
+**Tests**: `tests/util/metadata.test.ts` - Added deduplication test case
+
 ## Project Structure
 
 ```

package/guide/index.md

@@ -25,6 +25,7 @@ Protokoll transforms audio recordings into intelligent, context-enhanced transcr
 - [**Transcript Actions**](./action.md): Edit, combine, and manage transcripts
 - [**Feedback**](./feedback.md): Intelligent feedback for corrections
 
+
 ### AI Integration
 - [**MCP Integration**](./mcp-integration.md): Use Protokoll through AI assistants
 

package/guide/mcp-integration.md

@@ -152,16 +152,23 @@ If no `.protokoll` directory exists in the hierarchy, the AI will:
 |------|-------------|
 | `protokoll_add_person` | Add a new person to context |
 | `protokoll_add_project` | Add a new project with smart assistance |
-| `protokoll_suggest_project_metadata` | Generate project suggestions without creating |
-| `protokoll_update_project` | Update project by regenerating from source URL/file |
 | `protokoll_add_term` | Add a technical term with smart assistance |
-| `protokoll_suggest_term_metadata` | Generate term suggestions without creating |
-| `protokoll_update_term` | Update term by regenerating from source URL/file |
-| `protokoll_merge_terms` | Merge duplicate terms into one |
 | `protokoll_add_company` | Add a company |
-| `protokoll_add_ignored_term` | Add a term to ignore |
 | `protokoll_delete_entity` | Remove an entity |
 
+### Entity Editing Tools
+
+| Tool | Description |
+|------|-------------|
+| `protokoll_edit_person` | Edit person: add sounds_like variants, change role/company, etc. |
+| `protokoll_edit_term` | Edit term: add sounds_like variants, topics, projects, domain, etc. |
+| `protokoll_edit_project` | Edit project: add sounds_like variants, phrases, routing config, etc. |
+| `protokoll_update_term` | Regenerate term metadata from source URL/file (uses LLM) |
+| `protokoll_update_project` | Regenerate project metadata from source URL/file (uses LLM) |
+| `protokoll_merge_terms` | Merge duplicate terms into one |
+| `protokoll_suggest_project_metadata` | Generate project suggestions without creating |
+| `protokoll_suggest_term_metadata` | Generate term suggestions without creating |
+
 ### Transcript Tools
 
 | Tool | Description |
@@ -267,6 +274,26 @@ I also noticed 'WCNP' isn't in your terms. Would you like me to add it so future
 
 I found 3 meeting parts. Combined them into 'Full Meeting Notes' at `~/notes/client-alpha/2026-01/full-meeting-notes.md`. The original part files have been removed.
 
+### Editing Entities
+
+**User**: I need to add "Cartesian Time" as a sounds_like variant for the "cardigantime" term
+
+**AI**: *Uses the edit_entity prompt or calls `protokoll_edit_term` directly*
+
+I'll add that sounds_like variant for you.
+
+*Calls `protokoll_edit_term` with:*
+- id: "cardigantime"
+- add_sounds_like: ["Cartesian Time"]
+
+Done! I've added "Cartesian Time" as a phonetic variant for "cardigantime". Future transcripts will recognize this variant and correct it automatically.
+
+**User**: Can you also change Priya's company to "new-corp"?
+
+**AI**: *Calls `protokoll_edit_person`*
+
+Updated Priya's company from "acme-corp" to "new-corp".
+
 ## Best Practices
 
 ### 1. Set Up Project Configurations

package/guide/routing.md

@@ -66,6 +66,198 @@ routing:
 active: true
 ```
 
+## Classification Fields Reference
+
+Classification determines HOW Protokoll matches transcripts to projects. Each field serves a specific purpose in the routing algorithm.
+
+### explicit_phrases (Trigger Phrases)
+
+**Weight:** 90% (highest confidence)
+
+**Purpose:** Phrases that definitively indicate this project
+
+**When to use:** For unique project-specific phrases that rarely appear elsewhere
+
+```yaml
+classification:
+  explicit_phrases:
+    - "quarterly planning"
+    - "Q1 planning meeting"
+    - "roadmap review session"
+```
+
+**Example:** If transcript contains "This is a Q1 planning meeting", it routes to Quarterly Planning project with 90% confidence.
+
+**Manage:**
+```bash
+# CLI
+protokoll project edit quarterly-planning \
+  --add-phrase "Q2 planning" \
+  --remove-phrase "old phrase"
+
+# MCP
+use_mcp_tool('protokoll_edit_project', {
+  id: 'quarterly-planning',
+  add_explicit_phrases: ['Q2 planning'],
+  remove_explicit_phrases: ['old phrase']
+});
+```
+
+### topics
+
+**Weight:** 30% (lower confidence)
+
+**Purpose:** Theme keywords that suggest (but don't guarantee) this project
+
+**When to use:** For broad topic categories that help classify when combined with other signals
+
+```yaml
+classification:
+  topics:
+    - roadmap
+    - budget
+    - planning
+    - strategy
+```
+
+**Example:** Transcript mentioning "roadmap" and "budget" gets 60% confidence (30% + 30%).
+
+**Manage:**
+```bash
+# CLI
+protokoll project edit quarterly-planning \
+  --add-topic okrs \
+  --add-topic goals \
+  --remove-topic old-topic
+
+# MCP
+use_mcp_tool('protokoll_edit_project', {
+  id: 'quarterly-planning',
+  add_topics: ['okrs', 'goals'],
+  remove_topics: ['old-topic']
+});
+```
+
+### associated_people
+
+**Weight:** 60% (medium-high confidence)
+
+**Purpose:** Person IDs that indicate this project when mentioned
+
+**When to use:** When specific people are strongly tied to a project
+
+```yaml
+classification:
+  associated_people:
+    - priya-sharma     # Acme point of contact
+    - john-smith       # Project lead
+```
+
+**Example:** Transcript mentioning "Priya" routes to this project with 60% confidence.
+
+**Important:** Only associate people who STRONGLY indicate the project. If someone appears in many projects, don't associate them.
+
+**Manage:**
+```bash
+# CLI
+protokoll project edit client-alpha \
+  --add-person sarah-chen \
+  --remove-person old-contact
+
+# MCP
+use_mcp_tool('protokoll_edit_project', {
+  id: 'client-alpha',
+  add_associated_people: ['sarah-chen'],
+  remove_associated_people: ['old-contact']
+});
+```
+
+### associated_companies
+
+**Weight:** 60% (medium-high confidence)
+
+**Purpose:** Company IDs that indicate this project when mentioned
+
+**When to use:** For client projects or when company name definitively routes to a project
+
+```yaml
+classification:
+  associated_companies:
+    - acme-corp
+    - beta-industries
+```
+
+**Example:** Transcript mentioning "Acme Corp" routes to this project with 60% confidence.
+
+**Manage:**
+```bash
+# CLI
+protokoll project edit client-work \
+  --add-company acme-corp \
+  --add-company beta-industries
+
+# MCP
+use_mcp_tool('protokoll_edit_project', {
+  id: 'client-work',
+  add_associated_companies: ['acme-corp', 'beta-industries']
+});
+```
+
+### context_type
+
+**Weight:** Modifier (affects routing decisions)
+
+**Purpose:** Nature of content for this project
+
+**Options:**
+- `work` - Professional/business content
+- `personal` - Personal notes and ideas
+- `mixed` - Contains both work and personal
+
+```yaml
+classification:
+  context_type: work
+```
+
+**Manage:**
+```bash
+# CLI
+protokoll project edit my-project --context-type mixed
+
+# MCP
+use_mcp_tool('protokoll_edit_project', {
+  id: 'my-project',
+  contextType: 'mixed'
+});
+```
+
+## Viewing Classification
+
+Use `project show` to see all classification fields:
+
+```bash
+protokoll project show quarterly-planning
+```
+
+Output includes:
+- Context Type
+- Trigger Phrases (explicit_phrases)
+- Topics
+- Associated People (if any)
+- Associated Companies (if any)
+- Relationships (if configured)
+
+### Auto Tags
+
+The `auto_tags` field defines tags that are automatically added to transcripts routed to this project. These tags are combined with tags extracted from classification signals and deduplicated to ensure no duplicates appear in the final transcript metadata.
+
+```yaml
+routing:
+  auto_tags:
+    - work
+    - internal
+```
+
 ### Another Example
 
 ```yaml
@@ -173,6 +365,29 @@ Each signal contributes to a confidence score:
 
 The project with highest confidence wins.
 
+## Tags
+
+Tags are automatically generated from classification signals and added to transcript metadata:
+
+- **Extracted from signals**: Each classification signal's value (except `context_type`) becomes a tag
+- **Project auto_tags**: Tags defined in `routing.auto_tags` are added
+- **Automatic deduplication**: If the same tag appears from multiple sources (e.g., both as an `explicit_phrase` and as an `associated_project`), it appears only once in the final transcript
+
+Example: If "xenocline" is detected as both an explicit phrase and a project name, the transcript metadata will show:
+
+```yaml
+tags:
+  - xenocline
+```
+
+Not:
+
+```yaml
+tags:
+  - xenocline
+  - xenocline
+```
+
 ## API
 
 ### RoutingInstance

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@redaksjon/protokoll",
-  "version": "0.1.0",
+  "version": "0.3.0",
   "description": "Focused audio transcription with intelligent context integration",
   "main": "dist/main.js",
   "type": "module",