Skip to content

feat: Add Context Management Skill (Knowledge Layer) #167

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
mellanon wants to merge 121 commits into from
Closed

feat: Add Context Management Skill (Knowledge Layer) #167

mellanon wants to merge 121 commits into from

Conversation

@mellanon
Copy link

@mellanon mellanon commented Dec 8, 2025

Summary

Implements the Knowledge Layer from the Life OS vision — a complete pipeline for capturing, processing, and retrieving personal context.

Related Discussions:

What This Adds

Two CLIs (TypeScript/Bun)

CLI Purpose Key Commands
ingest Multi-modal capture via Telegram poll, watch, direct, search, load
obs Vault operations search, semantic, tags, read

Skill Definition

.claude/skills/Context/ with:

  • SKILL.md - Skill definition with workflow routing
  • README.md - Quick start guide
  • workflows/ - Semantic search and context loading workflows
  • docs/ - CONCEPTS, CLI-REFERENCE, CAPTURE-TIPS, SHORTCUTS

Documentation

Setup Guides:

Skill Documentation (.claude/skills/Context/):

Workflows:

iOS Shortcuts

Templates in shortcuts/ for clipboard capture and file sharing via Telegram.

Key Features

  • Multi-modal ingestion: Voice memos, photos, documents, URLs, text
  • Two-phase retrieval: search (discovery) → load (injection)
  • Inline hints: #tags @people /commands ~scope [metadata]
  • Fuzzy tag matching: Handles Whisper transcription errors gracefully
  • Privacy scopes: ~private / ~work separation
  • Archive pipeline: Structured naming with optional Dropbox sync

What's NOT Included (Yet)

Test Framework: A comprehensive 4-layer test pyramid exists (unit, integration, CLI, acceptance) but contains personal fixtures. A sanitized version with synthetic fixtures will follow in a separate PR.

Validated via Docker Clean Room

cd bin/ingest/deployment
make cleanroom-build SKILL_REPO=mellanon/Personal_AI_Infrastructure SKILL_BRANCH=release/context-skill
make cleanroom-test # ✅ PASSED## Required Configuration

Required

OBSIDIAN_VAULT_PATH=~/Documents/your_vault

For Telegram ingestion

TELEGRAM_BOT_TOKEN=your_bot_token
TELEGRAM_CHANNEL_ID=-100your_channel_id

Optional (semantic search)

OPENAI_API_KEY=sk-...## Follows PAI Principles

  • ✅ CLI-First - All operations via command line
  • ✅ Deterministic Code First - TypeScript/Bun, zero Python
  • ✅ Unix Philosophy - pbpaste | ingest direct --tags "project/pai"
  • ✅ Skills as Containers - Self-contained in .claude/skills/Context/

mellanon and others added 30 commits November 19, 2025 11:33
@mellanon
Ignore .env files
af1f6f0
@mellanon
Configure local PAI_DIR and upstream push safety docs
d56762e
@mellanon
Merge upstream/main into main
62e0844
@mellanon
Merge remote-tracking branch 'upstream/main'
29e9e5f
@mellanon
fix: update PAI_DIR path in settings.json for user-specific configura…
f94bccc 
…tion

- Changed PAI_DIR from a placeholder to a specific user path (/Users/andreas/.claude)
- Ensured DA variable is correctly set to "PAI"
- Maintained the comment for user guidance on setting PAI_DIR

This update enhances user experience by providing a clearer configuration example.
@mellanon @claude
feat: add context management system with obs and ingest CLI tools
0b64f94 
Two-stage ingestion pipeline:
- Stage 1: Telegram → Raw notes (#incoming tag)
- Stage 2: Raw → Processed notes (project tags, wisdom, etc.)

CLI tools:
- obs: vault operations (search, read, write, context, incoming, embed, semantic)
- ingest: Telegram ingestion (poll, process, status, profiles)

Key features:
- Project tags as context glue (obs context <project>)
- Configurable processing profiles (zettelkasten, simple)
- Automatic embedding updates via launchd
- Generic docs for upstream contribution

Docs:
- docs/architecture/context-system.md
- docs/architecture/telegram-ingestion.md
- docs/architecture/testing-and-migration.md
- docs/architecture/CONTRIBUTING.md

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix: handle Uint8Array blob from SQLite in semantic search
43ba6ad 
🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat: implement ingest processing pipeline
21ff8c8 
Processing pipeline now handles:
- voice/audio → ts (transcription) → fabric patterns → vault
- documents → marker (PDF/DOCX extraction) → fabric → vault
- URLs → fetch/yt → fabric → vault
- photos → tesseract OCR → vault
- text → direct to vault

Features:
- Content-aware routing (only transcribe audio, only OCR images)
- Profile-based fabric patterns per content type
- Paired output support (Raw + Wisdom notes)
- Automatic tag generation based on profile

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs: add detailed Telegram bot setup instructions
c928290 
🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat: add state tracking, security layer, and share-with-hints
742f0e5 
- Add SQLite state tracking for message processing (lib/state.ts)
  - Tracks pending/processing/completed/failed status
  - Enables idempotent reprocessing
  - Adds retry support for failed messages

- Add security layer for Telegram ingestion (lib/security.ts)
  - Prompt injection detection
  - Command whitelist validation
  - Rate limiting
  - Audit logging

- Add share-with-hints workflow
  - Extract #tags, @people, /commands from message text
  - Apply extracted metadata to note frontmatter
  - Document iOS Shortcut integration

- Add skills infrastructure
  - SKILL_CONTRACT.md documenting guarantees
  - self-test.ts health check for skills
  - context/ and vault/ skill directories

- Update ingest.ts with Telegram-valid reactions (👀👍👎🚫)
- Add test cases for state tracking (TEST-STATE-001 to 006)
- Document photo processing options for future Vision AI

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat: add Vision API, embedding config, iOS/macOS docs, and sanitize PII
e39d891 
- Add Vision API (GPT-4o) for photo processing with prompt-based analysis
- Make embedding model configurable (default: text-embedding-3-large)
- Add iOS/macOS integration test cases (Shortcuts, Quick Actions)
- Add PAI context integration test cases
- Add configuration layering docs for private fork customization
- Sanitize PII from all files for public contribution
- Add GitHub issue draft for RFC discussion
- Update launchd plist with placeholders
- Replace hardcoded names with generic examples

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat: add inbox filtering, AI title generation, and Events channel no…
bd9087e 
…tifications

- Filter messages to only process from PAI Inbox channel (isFromInbox)
- Add AI title generation for audio transcripts using gpt-4o-mini
- Add PAI Events channel for notifications with JSON payload
- JSON payload includes: event_type, status, severity, original_filename,
  output_files, output_paths for downstream routing (PagerDuty, etc.)
- Reply to processed messages with filename
- Add LaunchAgent plist and install script for daemon
- Lower AI title threshold from 100 to 20 chars

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
refactor: remove inbox replies, use Events channel only for notificat…
4bd7ee5 
…ions

- Remove replyWithObsidianLink calls from inbox channel
- PAI Inbox now stays clean (input only)
- PAI Events handles all notifications with JSON payload

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add mermaid syntax fix and photo AI titles
b90c63d 
- Add fixMermaidSyntax() to escape parentheses in mermaid node labels
- Auto-detect mermaid content in Vision API output and apply fix
- Add AI title generation for photos using gpt-4o-mini
- Improve filename sanitization (remove !@#$%^&()[]{}chars)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add pipeline v2 Phase 1 - metadata parsing and Jina URL…
b8179bd 
… extraction

Phase 1 of ingest pipeline v2 implementation:

- Add [key:value] metadata syntax parsing (source, device, user, type, category)
- Replace basic fetch with Jina AI Reader for clean markdown URL extraction
- Add fallback to basic fetch when Jina fails
- YouTube URLs continue to use yt tool (not Jina)
- Include source metadata in frontmatter output
- Add jinaApiKey and dropboxArchivePath to config
- Update self-test with v2 checks (Jina, Dropbox, metadata regex)
- Add ADR document: docs/architecture/ingest-pipeline-v2.md
- Add test cases TEST-INGv2-001 through TEST-INGv2-120

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add Phase 2 - archive pipeline, spoken hints, severity …
b40d701 
…icons

- Archive pipeline: /archive and /receipt commands with Dropbox sync
- Archive naming: {TYPE} - {YYYYMMDD} - {Description} - {CATEGORY}.{ext}
- Smart name preservation for pre-named files
- Spoken hints extraction from voice memos (Wispr Flow support)
  - "hashtag X" → #X, "at name" → @name, "forward slash X" → /X
- Merged hints: caption + spoken hints combined
- Notification severity icons: info/success/warning/error
- Updated skill contract with v2 capabilities
- Added spoken hints test cases (TEST-INGv2-130 to 134)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add Phase 3 - query command, AI intent parsing, enhance…
0a6f717 
…d self-test

Context Retrieval:
- Add `ingest query` command for searching vault and Dropbox archive
- Search by type (CONTRACT, RECEIPT), category (HOME, WORK), year
- Full-text search across archive filenames and vault content
- Results grouped by document type with source icons

AI Intent Parsing:
- LLM extracts pipeline, metadata, tags from natural language captions
- Uses gpt-4o-mini for fast, low-cost intent extraction
- 80% confidence threshold for automatic routing
- Falls back to default pipeline when confidence low

Self-Test Updates:
- Test 19: Events channel (TELEGRAM_OUTBOX_ID)
- Test 20: Spoken hints regex patterns
- Test 21: Archive naming pattern validation
- Test 22: whisper-cpp availability check
- Test 23: ffmpeg availability check

Updated skill contract and ADR with implementation status.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add query command and fix config loading
dd57246 
- Add executeQuery() for Telegram /query command
- Query supports semantic search, person mentions, project tags
- Fix obs config to load from both ~/.claude/.env and ~/.config/fabric/.env
- Fix embed.ts getOpenAIKey() to check .env files

Regression tests pass: self-test (22/22), security, retry, profiles

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add received_date, fix poll offset bug, add help command
52e6514 
- Add received_date to front matter (original Telegram message timestamp)
- Fix critical bug: poll command no longer advances offset (process only)
- Add /help command for Telegram bot with syntax documentation
- Add source_metadata to Events channel notifications
- Fix @FIRST Last mention pattern to normalize to first_last
- Change default notification severity from success to info
- Document regression tests in test-scripts.md

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add explicit fabric pattern commands and iOS docs
b025970 
- Add /summarize, /wisdom, /article, /meeting-notes commands
- Commands explicitly trigger fabric patterns (not auto based on content)
- Clips skip auto-patterns unless explicitly requested
- Clarify: use #1on1 as TAG, /meeting-notes as pattern COMMAND
- Update /help with pattern commands section
- Add iOS Shortcuts integration guide (docs/architecture/ios-shortcuts.md)
- Add regression tests for pattern commands (TEST-REG-015 to 020)
- Fix RTF document processing with HTML detection and debug logging

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add HTML-to-markdown conversion for iOS clipboard
768d0b6 
- Detect HTML in .txt files (both full documents and styled fragments)
- Convert to markdown with pandoc preserving links
- Clean up style attributes {style="..."} from pandoc output
- Fix bracket issues from nested HTML elements
- Extract actual URLs from tracking wrappers (tldrnewsletter, etc.)
- Strip UTM parameters for cleaner links

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs: add dictated photo intent to help, update test cases
a40da4b 
- Add dictated photo intent section to /help (any caption as Vision prompt)
- Add TEST-ING-045b for dictated mermaid intent
- Update iOS shortcuts documentation

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): complete archive pipeline with Dropbox sync and content…
c339bad 
… detection

- Add dictated intent detection for archive/receipt pipelines with type/category
- Detect document type (INVOICE, RECEIPT, CONTRACT, etc.) from Vision AI content
- Sync original files to Dropbox with proper naming convention
- Add Dropbox backlink in markdown frontmatter and content
- Skip fabric patterns for archive/receipt (straight-through processing)
- Add real-time embedding for new notes (fire-and-forget)
- Fix embedding model default to text-embedding-3-small (match existing DB)
- Update /help command with dictated intent examples
- Update SKILL_CONTRACT.md with new features
- Add test cases for archive pipeline and real-time embedding

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs: update help, skill contract, and test cases for archive pipeline
b88eced 
- Help: Add INVOICE, RECEIPT, BILL to auto-detected types
- Help: Note Vision AI detects type from content
- SKILL_CONTRACT: Add document type detection from content section
- SKILL_CONTRACT: Update archive naming examples with INVOICE
- test-scripts: Mark TEST-ARC-001-005 and TEST-EMB-001-002 as PASSED
- test-scripts: Add TEST-ARC-005 for Vision AI type detection
- test-scripts: Update version to 1.1.0

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(scope): add context separation with ~private/~work sigils
cd99739 
Security model: opt-in to work context
- New ingest defaults to scope/work (except archive/receipt → scope/private)
- Queries default to scope/work only (no tag = excluded)
- Support ~private/~work sigils and dictated hints
- Add --scope flag to obs search/semantic/context commands
- Existing vault stays private until explicitly tagged

Sigil syntax:
- ~private - personal/private content
- ~work - professional content (default for new notes)

Dictated detection:
- "scope private", "this is personal" → ~private
- "scope work", "for work" → ~work

Test cases: TEST-SCOPE-001 through TEST-SCOPE-024

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add document date hints for historic content
39d1521 
Add ability to specify historic document dates via:
- Structured metadata: [date:YYYY-MM-DD] or [date:DD/MM/YYYY]
- Dictated intent: "dated June 15th", "from last month", "from yesterday"

Document date is now:
- Used in archive naming (instead of processing date)
- Stored in frontmatter as document_date for ALL content types
- Available for voice memos, photos, documents, and all pipelines

This allows ingesting historical documents with accurate timestamps
reflecting when events actually occurred, not when they were digitized.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(ingest): use document date in note filename, not just frontmatter
5e9ae13 
When a document date is provided (via [date:...] or dictated intent),
the note filename now uses that date instead of today's date.

Example: "dated June 15th" creates "2024-06-15 - Invoice.md"
instead of "2024-12-02 - Invoice.md"

This applies to ALL content types, ensuring historical documents
have filenames reflecting their actual date.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs: update test cases and self-test for scope and date features
4883ceb 
- Add TEST-DATE-020/021/022 for note filename using document date
- Update skill contract to document note filename uses document date
- Add scope sigil pattern test (Test 25) to self-test
- Add document date pattern test (Test 26) to self-test

All 26 self-tests pass.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): add comprehensive test framework with fixtures
cb5ae91 
- 45 unit tests (fixture-based) covering scope, date, archive, regression
- Integration test support for isolated test channels
- Test channel config (TEST_TELEGRAM_CHANNEL_ID/OUTBOX_ID)
- Fixture capture and validation framework
- Test assets (PDFs, images, documents) for media tests

Unit tests run via `ingest test run` (~3 min)
Integration tests require user-sent messages to test channel

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(ingest): use two-thread approach for integration tests
361ff70 
Telegram's getUpdates requires active polling when messages arrive.
Changed processViaIngest() to start `ingest watch` in background
before messages are sent, then poll for vault file creation.

This mirrors production where the watch daemon runs continuously.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
mellanon and others added 26 commits December 9, 2025 08:08
@mellanon @claude
feat(shortcuts): Update device details to use Device Model
9d26da7 
- clipboard-capture: Add Mac conditional for caption prompt
- file-capture: Change Device Name to Device Model
- Both templates now use consistent device identification

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(shortcuts): Add universal caption prompt and reorder actions
76efb3c 
- clipboard-capture: Add caption prompt for all devices before Mac conditional
- file-capture: Reorder to set device variable before asking for caption
- Both shortcuts now prompt for caption on all platforms

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
refactor: Update contribution workflow and fix shortcut
351b540 
- Replace pai-contrib with public fork workflow in development plan
- Remove Mac conditional from clipboard-capture (single universal caption)
- Simplify architecture diagram for standard GitHub fork workflow

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(context): Add structured tag taxonomy for auto-tagging
809d800 
- Add tags.json with DailyFeed, incoming, raw, wisdom, meeting-notes, etc.
- Include usage guidelines, auto-detect keywords, and examples
- Define project keywords for pai, security, architecture
- Add source and topic tag categories

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): Add tags to PAI event notifications
2854e27 
- Add tags field to NotificationOptions interface
- Include tags in event payload when present
- Pass tags from processed content to notifications

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): Add taxonomy-based keyword auto-tagging
7f997f0 
- Create taxonomy.ts to load tags.json and match keywords
- Add fuzzy matching for transcription errors (Levenshtein ≤2)
- Integrate keyword matching before AI tagging in process.ts
- Skip AI if keyword coverage is sufficient (≥3 semantic tags)
- Update AI prompt to strongly prefer taxonomy tags
- Rename DailyFeed → feed for brevity

Tag priority: hints → keywords → AI → status tags

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs: Add 3-repo workflow documentation and sanitization script
12587f2 
- Update CLAUDE.md with full repo architecture and workflows
- Create sanitize-for-contrib.sh to check for personal data before fork push
- Simplify context-skill-development-plan.md to quick reference format
- Document personal vs generic file locations

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(shortcuts): Add caption truncation and update sanitization script
740f405 
- Shortcut templates now truncate caption to 1020 chars (Telegram limit)
- Fixed sanitize-for-contrib.sh bash array compatibility
- Added explicit no-auto-commit rules to CLAUDE.md for fork safety

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): Add /tag command and simplify tagging to user+AI only
16a1efe 
- Remove keyword-based auto-tagging (was too aggressive with project tags)
- Add /tag command to force AI tagging even with 3+ user tags
- AI tagging now runs if: user provides <3 tags OR /tag command used
- Make URL fetching opt-in via /article, /wisdom, /summarize, /fetch
- URLs saved as-is by default (no auto-fetch)
- Update /help with new AI Tagging section and /tag command

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): Add sender bot token and dynamic device detection
58f1bb3 
- Add TELEGRAM_SENDER_BOT_TOKEN for sending to inbox (aligns with iOS shortcuts)
- TELEGRAM_BOT_TOKEN now used only for polling/reading
- sendToInbox, sendPhotoToInbox, sendDocumentToInbox use sender token
- Add getDeviceModel() for dynamic device detection via system_profiler
- Normalize metadata to lowercase: [device:mac studio] [user:andreas]
- Add Jira skill (Commitment domain) with CLI-first architecture

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(context): Add URL capture workflow with extract_page_content pat…
58b335e 
…tern

- Add /extract-content and /clean pipeline commands for web content extraction
- Create extract_page_content fabric pattern for stripping web UI noise
- Update Context SKILL.md with URL capture workflow using Fabric + pipeline

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(ingest): Add extract-content command to security whitelist
6a33884 
- Add /extract-content and /clean to allowed commands in security.ts
- Fix SKILL.md URL capture workflow to use --pipeline flag

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(ingest): Use Bun.spawn for fabric patterns to fix shell escaping
9d51074 
Replace shell echo with Bun.spawn stdin for applyFabricPatterns().
Complex markdown with code blocks, backticks, and quotes was breaking
shell escaping. Now content is piped directly to fabric via stdin.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(ingest): Add bidirectional links and tag propagation for paired …
00304cd 
…notes

- Wisdom notes now inherit user-provided tags from ingestion command
- Add hasWisdomSibling flag to ProcessedContent interface
- Raw notes link forward to Wisdom: 📝 **Processed:** [[...-Wisdom]]
- Wisdom notes link back to Raw: 📄 **Source:** [[...-Raw]]

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(obs): Improve search tag display and fix inline tag extraction
fa830e5 
- Filter system/meta tags from display (incoming, raw, source/telegram)
- Show user-meaningful tags first (architecture, philosophy, etc.)
- Fix inline tag regex to exclude markdown anchors like #-updates
- Increase tag display limit from 2 to 3 tags, 30 to 35 chars

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(obs,ingest): Use frontmatter tags for type detection, restrict ar…
2e42400 
…chive metadata

- obs search Type column now uses frontmatter tags only (not inline content tags)
- Add frontmatterTags field to ParsedNote and SearchResult interfaces
- Restrict document_type/category/date to archive pipeline only
- Prevents false type detection from documentation content

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(obs): Add tag add/remove command for deterministic tag management
f54a8b4 
Adds `obs tag add/remove` command to manage note tags via CLI:
- obs tag add 3 architecture (by index from last search)
- obs tag add "Note Name" project/pai (by note name)
- obs tag remove 5 incoming (remove tag)

Supports note selectors: numeric index, note name, or full path.
Updates SKILL.md with command reference and examples.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(obs): Add --tag filter to load, --type filter to search
3f30d16 
- obs load --tag <tag>: Filter loaded results by tag (AND logic)
- obs search --type <type>: Filter search by type (transcript, wisdom, etc.)
- Updated SKILL.md with new filter examples

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(obs): Add --any-tag flag for OR logic in tag filtering
ae812eb 
Complements existing --tag (AND) with --any-tag (OR) for flexible queries:
- --tag a --tag b: notes with BOTH tags (intersection)
- --any-tag a --any-tag b: notes with ANY tag (union)

Added to search, semantic, and load commands. SKILL.md updated with
language pattern guide for when to use AND vs OR.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
fix(ingest): Filter metadata values from AI-suggested tags
dd9eb67 
Prevents metadata pollution where values from [source:X][device:Y][user:Z]
markers were being suggested as tags by the AI tagger.

- Filter metadata values (source, device, user) from AI tag suggestions
- Add metadata pollution patterns to obs display filter for existing notes

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs(context): Add explicit table format with Tags column
f135144 
Provides concrete example of expected search results table format
to ensure Claude includes Tags column and full load options.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
feat(obs): Multi-line tags and wider columns in search results
222e391 
- Title column: 40 → 50 chars
- Tags: Show ALL tags (removed 3-tag limit)
- Tags wrap to multiple lines with proper table alignment
- Tag line width: 50 chars before wrapping

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
refactor(skills): Trust Claude Code native skill discovery for Context
1e015c6 
- Add ${PAI_DIR} variable substitution in load-core-context.ts hook
- Remove Context skill routing from CORE (auto-discovered by Claude Code)
- Context skill now discovered natively from ~/.claude/skills/Context/

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
docs(skills): Add Context skill routing with extended docs reference
dfc1c20 
- Add Context skill routing back to CORE with two-phase workflow reminder
- Reference extended Context docs (workflows, CLI reference, concepts)
- Update Context description with better trigger examples

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon @claude
chore: Sanitize personal data in code comments and examples
7050a53 
- Replace vault path in SKILL.md example with ${OBSIDIAN_VAULT_PATH}
- Add tags.example.json as template, gitignore personal tags.json
- Replace [user:andreas] with [user:testuser] in code comments and tests

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
@mellanon
chore: Sanitize personal paths in self-test.ts
b77c896
@mellanon mellanon force-pushed the release/context-skill branch from eab4a77 to b77c896 Compare December 10, 2025 10:05
@mellanon mellanon closed this Dec 10, 2025
@mellanon
Copy link
Author

mellanon commented Dec 10, 2025

Apologies... I had to close this PR. My local git setup got tangled and couldn't figure out a way to rebase the PR with my latest work. I'll create a new one, once I have untangled my setup.

@mellanon mellanon deleted the release/context-skill branch December 10, 2025 11:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@mellanon