vault backup: 2026-04-05 15:37:59

.claude/saves/index.md

@@ -0,0 +1,5 @@
+# Save Index
+
+| ID | Date | Type | Title | Tags |
+|----|------|------|-------|------|
+| SAVE_20260404_001 | 2026-04-04 | research | Obsidian document archiving via OpenClaw agent | obsidian, openclaw, telegram, archiving, pdf |

.claude/saves/research/SAVE_20260404_001_obsidian-openclaw-archiving.md

@@ -0,0 +1,59 @@
+# Save Entry: Obsidian Document Archiving via OpenClaw Agent
+
+**Date**: 2026-04-04
+**Type**: research
+**Tags**: obsidian, openclaw, telegram, archiving, pdf, ocr, skills
+
+## Context
+- **What**: Designing a workflow to archive documents (PDFs, images) into the Obsidian vault using an OpenClaw agent on a Razer Blade laptop
+- **Why**: User wants to send documents via Telegram, have the agent convert them to markdown, and save both originals + markdown into the vault
+- **Where**: `/Users/ylu/Documents/me/obsidian-yanxin` (git-synced between Mac and Razer Blade)
+
+## Vault Structure
+- 179 markdown files
+- Key folders: `copilot/`, `documents/` (has `argo/` subfolder), `journals/`, `notes/` (books, martial_arts, medical), `Readwise/`, `templates/`, `travels/`, `writing/`
+- Git-synced repo, master branch
+
+## Key Findings
+
+### OpenClaw Capabilities (from docs.openclaw.ai)
+1. **Telegram integration** is production-ready with media support (PDFs, images, audio)
+   - Supports long polling (default) and webhook mode
+   - Media files up to 100MB (`mediaMaxMb`)
+   - Inbound media normalized into shared channel envelope
+2. **PDF tool** built-in: extracts text from PDFs, supports local paths and URLs, native mode for Anthropic/Google providers
+3. **Vision** available for image OCR/transcription
+4. **Skills system**: AgentSkills-compatible folders with `SKILL.md` files
+   - Skill locations (priority order): workspace skills > project agent skills > personal agent skills > managed skills > bundled > extra dirs
+   - Install custom skills in `~/.openclaw/skills/<skill-name>/SKILL.md` or workspace `skills/` folder
+   - Browse/install from clawhub.ai
+5. **Slash commands** supported for triggering skills
+
+### Proposed Architecture
+- **No inbox folder needed** - Telegram IS the inbox
+- User sends PDF/image via Telegram to the OpenClaw agent
+- Agent receives media, processes it (PDF tool or vision), converts to markdown
+- Agent saves original + .md to `documents/archive/` in the vault
+- Agent runs `git add`, `commit`, `push`
+- User pulls on Mac to get the files
+
+### What Needs to Be Built
+- A custom OpenClaw skill (e.g., `/archive`) in `~/.openclaw/skills/archive-to-vault/SKILL.md`
+- The skill should:
+  1. Receive media file from Telegram message
+  2. Use PDF tool or vision to extract content to markdown
+  3. Save both files to `documents/archive/` in the vault
+  4. Git commit and push
+
+## Next Steps
+1. Look at OpenClaw skill SKILL.md format/examples more closely
+2. Scaffold the `archive-to-vault` skill
+3. Configure git on the Razer Blade for auto-commit/push
+4. Test with a sample PDF and image via Telegram
+
+## References
+- Docs: https://docs.openclaw.ai/
+- Telegram channel docs: https://docs.openclaw.ai/channels/telegram.md
+- Skills docs: https://docs.openclaw.ai/skills/
+- PDF tool docs: https://docs.openclaw.ai/tools/pdf.md
+- Vault repo: `/Users/ylu/Documents/me/obsidian-yanxin`