18 KiB
18 KiB
| pageType | id | title | sourceType | sourcePath | ingestedAt | updatedAt | status |
|---|---|---|---|---|---|---|---|
| source | source.all-sources | all-sources | local-file | /home/owner/.openclaw/workspace/MEMORY.md | 2026-06-26T14:04:08.918Z | 2026-06-26T14:04:08.918Z | active |
all-sources
Source
- Type:
local-file - Path:
/home/owner/.openclaw/workspace/MEMORY.md - Bytes: 17902
- Updated: 2026-06-26T14:04:08.918Z
Content
### Telegram Bot — configuration (04.06.2026)
- **Name:** Cl
- **Bot ID:** 8788956736
- **Username:** @clawpi_01_bot
- **User Chat ID:** 862485388
### Network Fix
- **Problem:** Telegram API resolves to IPv6 (2001:67c:4e8:f004::9), Pi has no IPv6 routing
- **Fix:** export NODE_OPTIONS="--dns-result-order=ipv4first" added to /etc/environment
- **Gateway restart:** sudo systemctl --user restart openclaw-gateway
- **Status:** ✅ fixed
### Memory Search Configuration (2026-06-17)
- **enabled:** true
- **provider:** ollama (ollama-local)
- **model:** bge-m3
- **remote.baseUrl:** http://192.168.1.200:11434
- **embedding vector dims:** 1024
- **storage:** ~/.openclaw/memory/{agent}.sqlite (main, coder-agent, evaluator-agent, humanizer-agent, setup-agent)
- **index:** 1 file / 8 chunks indexed per agent (MEMORY.md source)
- **all agents:** Embeddings ready, Vector store ready, Semantic vectors ready ✅
- **Note:** OpenClaw CLI `openclaw memory index` runs on Pi but embedding API needs Dell Precision Ollama (192.168.1.200:11434). Gateway process uses config baseUrl correctly.
### AI Models (2026-06-22)
- **agents.defaults.model.primary:** `ollama-local/qwen3.6-35b-192k`
- **agent.models:**
- main → inherits default: `ollama-local/qwen3.6-35b-192k`
- coder-agent → `ollama-local/qwen3-coder:30b` (unchanged — specialized)
- evaluator-agent → `ollama-local/qwen3.6-35b-192k` (changed from `qwen3.6-35b-128k`)
- humanizer-agent → `ollama-local/qwen3.6-35b-192k` (changed from `qwen3.6-35b-128k`)
- setup-agent → `ollama-local/qwen3.6-35b-192k` (unchanged)
- **imageModel.defaults.primary:** `ollama-local/qwen3-vl:32b` (changed 2026-06-22; qwen2.5vl broken by CLIP bug in Ollama 0.30.8)
- **memorySearch.model:** `bge-m3` (embedding, not generative)
- **dreaming model:** `ollama-local/qwen3.6-35b-128k` (separate plugin config)
### Discord Bot (2026-06-21) — ⚠️ BLOCKED
- **Token:** configured in openclaw.json channels.discord.token
- **Guild:** 1518244752967864330
- **Plugin:** enabled (discord)
- **Skill:** enabled (discord)
- **Problem:** Raspberry Pi cannot reach Discord API — connection timeout to 162.159.x.x subnet
- **Reason:** ISP/router blocks Discord traffic from home network
- **Fix needed:** Run bot on VPS or fix DNS/routing for Discord
- **Status:** ⚠️ blocked by network
### Systemd Unit — /etc/environment binding (2026-06-12)
- **Problem:** systemd does not read /etc/environment — it's a PAM file, triggered upon login, not for systemd services.
- **Fix:** Added `EnvironmentFile=/etc/environment` to the [Service] section of the openclaw-gateway.service unit.
- **Command:** `systemctl --user edit --force --full openclaw-gateway.service` (created a new unit).
- **Then:** `systemctl --user daemon-reload` + `systemctl --user restart openclaw-gateway`
- **Result:** All vars from /etc/environment are visible to the gateway process.
## Search Tool Priority
Search priority:
web_search → Tavily Search → Tavily Extract → Firecrawl Search/Scrape.
Use:
- web_search for quick lookup;
- Tavily for high-quality search, news, time/domain filtering;
- Tavily Extract for article/page extraction;
- Firecrawl only for JS-heavy or difficult pages;
- web_fetch only for simple URL retrieval.
Always choose the lightest tool that can solve the task.
## Configuration Management
### Files
- **CHANGELOG.md** — change history. Each significant infrastructure change → record (created 2026-06-11).
- **MEMORY.md** — persistent system state (infrastructure, models, bot, settings). Not history, but final state.
- **AGENTS.md** — routing and orchestration policy. Does not contain settings.
- **Wiki vault** (`/home/owner/obsidian-vault`) — structured entities from MEMORY.md + CHANGELOG.md.
### Vault state (2026-06-21 06:04)
- memory-wiki: vaultMode=bridge, ingest.autoCompile=true, render.createDashboards=true, search.backend=shared, search.corpus=all, context.includeCompiledDigestPrompt=true
- memory-core: /home/owner/.openclaw/workspace/memory-core/ (daily/, dreams/, inbox/, index.md)
- dreaming: enabled=true, frequency=daily, model=ollama-local/qwen3.6-35b-128k, timezone=Europe/Moscow
- wiki: 18 sources, 5 entities, 3 concepts, 2 syntheses, 10 reports — compiled to 38 pages
- wiki lint: 118 issues (22 errors, 96 warnings), 0 contradictions, 10 provenance gaps
- last ingest: 2026-06-25 00:32 MSK (MEMORY.md)
- wiki_apply remains primary mechanism for creating/updating structured content
- known bug: bridge import syncs 0 artifacts (GH issue #65698) — manual wiki ingest + wiki_apply needed
### Heartbeat
- **Interval:** 5m
- **Config:** agents.list[0].heartbeat.every = "15m" in openclaw.json
### Cron Jobs — Mail Check (2026-06-14)
- **check-inbox-5m:** УДАЛЁН (удалён 20:17)
- **check-email-daily:** новый cron, каждый день в 9:00 МСК (`0 9 * * *`, Europe/Moscow), delivery → telegram:862485388
- **Prompt:** "Проверь inbox для arno@eurostela.ru через /home/owner/.openclaw/workspace/skills/nextcloud-aio-oc/scripts/mail.py. Посмотри последние 3 письма. Если есть новые — покажи тему и отправителя." (timeout: 120s)
- **Status:** ✅ работает
### Cron Jobs (routing policy reminder — 2026-06-12)
- **File:** /etc/sudoers.d/owner
- **Content:** `owner ALL=(ALL) NOPASSWD: ***`
- **Permissions:** 440
- **Purpose:** setup-agent needs root access for systemd, /etc/, packages, networking
- **Agent access:** setup-agent has passwordless sudo; main-agent has no sudo access
- **Audit trail:** all infra changes logged in CHANGELOG.md + MEMORY.md
### Agent Boundaries
- **main-agent:** routing, planning, coordination, status reporting only
- **setup-agent:** infrastructure (systemd, /etc/, packages, networking, firewall) — has NOPASSWD sudo
- **coder-agent:** implementation, code, scripting
- **evaluator-agent:** validation, audit, correctness checks
- **humanizer-agent:** text rewriting, tone, readability
- **No agent may self-certify changes** — evaluator validates all infra mutations
- **All infra changes require CHANGELOG.md entry + MEMORY.md update + wiki compile**
### Change workflow
- Infra/config changes → delegate to setup-agent (per AGENTS.md routing)
- Agent MUST update CHANGELOG.md after infra changes
- Agent MUST update MEMORY.md with final state only (not history)
- All files loaded at session start via project context
- No agent may self-certify changes; evaluator-agent validates infra mutations
### Wiki Ingestion (event-driven, per AGENTS.md §16)
- Main-agent checks wiki sources before every reply
- If sources changed → auto-spawn setup-agent subagent for wiki ingest + apply + compile + lint
- Sources: USER.md, MEMORY.md, AGENTS.md, SOUL.md, IDENTITY.md, TOOLS.md, EMBEDDING-TRADING.md
- No cron — purely event-driven
- Setup-agent MUST create/update entities, concepts, syntheses from sources
- Setup-agent MUST update CHANGELOG.md after wiki mutations
### Cron Jobs (routing policy reminder — 2026-06-12)
- **CRITICAL:** Cron config changes = infrastructure task → delegate to setup-agent (AGENTS.md §5.1)
- Main-agent MUST NOT modify cron jobs directly
- Violation example: main-agent updated check-dreaming delivery manually on 2026-06-12
- Future cron changes → setup-agent only
## Agent Behavior Rules
### Automatic Cron for Deferred Checks (2026-06-11)
- If the agent says "I'll check/look at/evaluate/monitor in N" — a cron job is immediately created for the required time.
- Don't ask the user, don't discuss it — just create and report.
- Report: "Done, cron job for [time]"
### Rule Creation (2026-06-11)
- If the agent decides that a new rule needs to be created or an existing one modified, it formulates the rule and requests the user's permission to write it.
- Does not make changes to MEMORY.md / AGENTS.md without the user's explicit consent.
- Upon approval, it writes the final state to MEMORY.md and updates CHANGELOG.md.
## Email Addresses
- **Arno (Agent):** arno@eurostela.ru (via Beget)
- **Sergey (Owner):** sergey@eurostela.ru
## Nextcloud Mail
- **Account:** arno@eurostela.ru
- **IMAP:** imap.beget.com:993 (SSL)
- **SMTP:** smtp.beget.com:465 (SSL)
- **Script:** /home/owner/.openclaw/workspace/skills/nextcloud-aio-oc/scripts/mail.py
- **Capabilities:** inbox, get, send, reply, folders, search, delete
- **Status:** ✅ connected
- **Password:** stored in /etc/environment (NEXTCLOUD_MAIL_PASS)
### Email Behavior Rule (2026-06-17) (v2.0)
**1. Deleting a specific message (by UID):**
- Perform an IMAP MOVE (UID MOVE) from the INBOX to the Trash folder.
- DO NOT use the '\Deleted' flag + EXPUNGE (to avoid permanent deletion).
- After the MOVE, verify that the message is no longer in the INBOX (by UID).
**2. Clearing the INBOX (delete all messages):**
- MOVE all UIDs from the INBOX to the Trash (in batches of 500-1000 messages).
- After completion, verify that the INBOX status (EXISTS) = 0.
- In case of an error, roll back (do not delete partially).
**3. Clearing the Trash:**
- First, ask the user for confirmation (dangerous action).
- Upon confirmation: select the Trash folder, set the '\Deleted' flag on ALL messages, and perform an EXPUNGE (or UID STORE + Gmail-style delete). - After the operation, check that the Trash is empty (EXISTS = 0).
- Log this action (number of deleted emails).
### Email Script Fix (2026-06-17)
- `mail.py` — fixed `\Deleted` escape in IMAP STORE command; added `--all` support for `cmd_delete`.
## Nextcloud AIO (2026-06-14)
- **Skill:** nextcloud-aio-oc (installed in workspace/skills/nextcloud-aio-oc)
- **URL:** https://datacloudport.ru
- **User:** Arno
- **Auth:** Basic Auth (NEXTCLOUD TOKEN account password)
- **Timezone:** Europe/Moscow
- **Status:** ✅ connected — OCS API + Notes list + WebDAV all working
- **Env:** /etc/environment (NEXTCLOUD_URL, NEXTCLOUD_USER, NEXTCLOUD_TOKEN, NEXTCLOUD_TIMEZONE)
- **Capabilities:** Notes, Tasks, Calendar (CalDAV), Files (WebDAV), Contacts (CardDAV)
- **Office libraries:** python-docx 0.8.11, openpyxl 3.1.5, python-pptx 1.0.2, reportlab 4.4.4, pypdf 5.6.0
- Fully supports .docx/.xlsx/.pptx files (read/edit/create)
- PDF reading and editing
- **Contacts:** 13 entries (8 user, 5 system)
- **OCS fix (2026-06-14):** OCS API returns 412 without `OCS-APIRequest: true` header (Nextcloud 30+ CSRF protection). Added to bundled request() function in nextcloud.js — all OCS endpoints now return 200. WebDAV unaffected.
- **Warning:** fix applied to compiled bundle; will be lost if skill is updated or reinstalled.
### Nextcloud Behavior Rule (2026-06-15)
- When a user requests an action with "Nextcloud" without specifying an account, immediately use Arno's account on datacloudport.ru without asking for confirmation or clarification.
### WordPress — Eurostela.ru (2026-06-15)
- **URL:** https://eurostela.ru
- **Тип:** WordPress (3+ года), Elementor + ACF
- **Пользователь:** Arno, role: editor
- **Application Password:** arno — LwZFmiWyCrhPHlUwAShq7qFD (имя: Arno-API)
- **API endpoint:** https://eurostela.ru/wp-json/wp/v2/
- **Статус:** ✅ Connected, REST API is working, authorization completed
- **Категория по умолчанию:** "User information" (id=100)
- **Назначение:** content, SEO optimization, website promotion
### WooCommerce REST API — Eurostela.ru (2026-06-24)
- **API base:** https://eurostela.ru/wp-json/wc/v3/
- **Consumer Key:** `ck_6ae9d729f33e1672ffed8bd41c042b14db74aabf`
- **Consumer Secret:** в окружении `WOOCOMMERCE_API_SECRET` (не разглашать)
- **Auth:** Basic Auth (key:secret) — обязательны оба ключа
- **Пример запроса:**
- Поиск по SKU: `GET /wc/v3/products?sku=MR-038`
- Список товаров: `GET /wc/v3/products?per_page=10&orderby=date&order=desc`
- Один товар: `GET /wc/v3/products/{id}`
- **SKU форматы:** кириллические (МР-xxx) и латинские (MR/MP-xxx) — разные коды
- **Статус:** ✅ API keys working, read access confirmed
## Project Creation and Storage Policy
Reference:
/home/owner/.openclaw/workspace/project-rules.md
### Project Registry
All projects are registered in:
/home/owner/.openclaw/workspace/projects/index.md
### Project Statuses
Allowed values:
planning
active
paused
completed
archived
### Required Project Files
Every project must contain:
```text
PROJECT.md
README.md
CHANGELOG.md
```
Purpose:
```text
PROJECT.md → current project state
README.md → human-readable overview
CHANGELOG.md → project history
```
### Required Project Directories
```text
tasks/
decisions/
audit/
reports/
assets/
content/
docs/
notes/
contracts/
tmp/
archive/
```
### Project Statuses
Allowed values:
```text
planning
active
paused
completed
archived
```
### Project Registry
All projects tracked in:
```text
/home/owner/.openclaw/workspace/projects/index.md
```
### Context Loading Policy
Before working on a project:
1. Read PROJECT.md
2. Read open tasks
3. Read latest CHANGELOG entries
4. Read latest notes
5. Load additional files only if required
Never load the entire project by default.
### Storage Policy
Primary storage:
```text
workspace/projects/
```
Backup:
```text
Nextcloud /Projects/
```
Sync direction:
```text
local → Nextcloud
```
Automatic reverse synchronization is prohibited.
### Workspace File Policy
- **Корень `/workspace/`** — только файлы конфигурации OpenClaw (AGENTS.md, MEMORY.md, SOUL.md и т.д.) и папки `projects/`, `skills/`, `memory-core/`.
- **Запрещено**: создавать файлы вне папок проектов и skills. Все создаваемые файлы должны идти внутрь `/projects/<project>/` или `/skills/`.
- **При нарушении** — автоматически перенести файл в соответствующую папку проекта.
### Research and Decisions
Research → facts, tests, evaluations
Decisions → approved choices
Never mix them.
Research location:
notes/research/
Decision location:
decisions/
### Project Registry
See:
/home/owner/.openclaw/workspace/projects/index.md
### Agent Boundaries
Agents may:
- create project files
- modify project files
- update PROJECT.md
- update CHANGELOG.md
- create reports
- create decisions
- create tasks
Agents may not:
- modify archived projects
- delete archives
- access unrelated projects without permission
- modify system configuration without approval
- move files between projects without approval
### Project Versioning Policy (2026-06-20)
#### Scope
- **All regular files** inside `workspace/projects/<project-name>/`, at any nesting level.
- Excludes:
- Cache/temporary files (`.DS_Store`, `Thumbs.db`)
- Hidden VCS directories (`.git`)
- The `archive/` directories themselves (not recursively versioned)
- Any directory (only regular files are versioned)
#### Core idea
- Active (current) file stays **in place** with its original name — **no revision suffix**.
- Every edit creates a **dated, numbered snapshot** in a **local `archive/` subfolder** located **in the same directory** as the file being edited.
- Archive filenames preserve the original filename **and extension** to avoid collisions: `<filename>.<ext>-<YYYY-MM-DD>-BU_<N>.<ext>`.
- The archive/ folder is shared by all files in the same directory. It is never created as a subfolder of a file.
- Revision names always preserve the original file extension.
#### Pattern per file:
```
/home/owner/.openclaw/workspace/projects/<project-name>/content/
├── article.md ← active file
├── notes.md ← active file
├── articleassets.md ← active file
└── archive/
├── article-2026-06-20-BU_1.md
├── notes-2026-06-20-BU_1.md
├── article-2026-06-20-BU_2.md
└── notes-2026-06-22-BU_2.md
```
#### Behavior on edit
##### First edit (no archive exists for this file)
- In the same folder as `<file>.md`, create `archive/` if it doesn't exist.
- Copy the current content of `<filename>.<ext>` into `archive/<filename>-<YYYY-MM-DD>-BU_1.<ext>`.
- Keep editing the original `<filename>.<ext>` — it remains the active version.
(No data loss, no empty active file.)
##### Subsequent edits
- Ensure the local `archive/` folder exists.
- Find all revisions for this file in that `archive/` folder (by `<filename>-*.<ext>` pattern) and determine the highest revision number `N`.
- Copy the current active `<filename>.<ext>` into `archive/<filename>-<current-date>-BU_<N+1>.<ext>`.
- Continue editing `<filename>.<ext>` as the active version.
##### Important notes
- The **active file is never moved** — only **copied** into the archive.
This guarantees the working copy is always ready and unchanged.
- Revision names include the original filename (e.g. `readme-2026-06-20-BU_1.md`), ensuring unique identification even if multiple `.md` files share the same `archive/` folder.
- There is no separate "base" file after the first edit — the original file remains the single active version.
- The archive/ directory appears once per working directory (e.g., assets/archive/, content/archive/), not once per file.
#### Relationship with storage/backup policy
- Archive revisions are part of the project and **are pushed to Nextcloud** together with all other files during manual sync or at natural checkpoints (task completion, milestone).
- No automatic background sync of archives.
- See `Storage Policy (2026-06-16)` for overall backup rules.
## TEST-2026-06-21
Raspberry Pi hostname is CLAWPI.
openclaw memory search уникальный_тест_12345
Notes
Related
- No related pages yet.