docs: README follows the beaver_agent layout, MODELS.md moves to the vault
This commit is contained in:
@@ -1,65 +0,0 @@
|
|||||||
# Модели: как вводить новую (чеклист)
|
|
||||||
|
|
||||||
Единственное место с процедурой. Проверено 2026-09-01 боевым тестом на dell.
|
|
||||||
|
|
||||||
## Главный инвариант
|
|
||||||
|
|
||||||
**API отбрасывает новые модели по версии клиента.** `claude` CLI 2.1.248 с
|
|
||||||
`--model claude-fable-5-1` получает от API
|
|
||||||
`400: version 2.1.251 or newer is required` — id проходит насквозь, но сервер
|
|
||||||
режет по минимальной версии Claude Code для этой модели. Поэтому порядок
|
|
||||||
строго такой: **сначала CLI, потом id в конфигах.**
|
|
||||||
|
|
||||||
CLI живёт в двух местах:
|
|
||||||
|
|
||||||
- **гейтвей (контейнер на dell)** — забандлен в питоновский пакет
|
|
||||||
`claude-agent-sdk` (`_bundled/claude`), замораживается при сборке образа.
|
|
||||||
Версия CLI = версия пакета, соответствие в
|
|
||||||
[CHANGELOG SDK](https://github.com/anthropics/claude-agent-sdk-python/blob/main/CHANGELOG.md)
|
|
||||||
(например 0.2.146 → CLI 2.1.248, 0.2.150 → 2.1.257).
|
|
||||||
- **t3code на машинах (mac, dell-хост)** — обычный `claude` пользователя
|
|
||||||
(`~/.local/share/claude/versions/…`, `/usr/local/bin/claude`), сам
|
|
||||||
автообновляется; вручную — `claude update`.
|
|
||||||
|
|
||||||
## Чеклист «вышла новая модель X»
|
|
||||||
|
|
||||||
1. **beaver-gateway**: поднять SDK до версии с достаточно новым CLI:
|
|
||||||
`pyproject.toml` (`claude-agent-sdk>=…`) + `uv lock -P claude-agent-sdk`,
|
|
||||||
закоммитить, запушить `main`.
|
|
||||||
2. **beaver-agent/config.py** (строки ~383–399, список агентов): поменять id
|
|
||||||
моделей у `deep`/`dispatcher`/… Effort там же.
|
|
||||||
3. **beaver-agent/t3code.toml** (дефолты t3code-mcp per-machine,
|
|
||||||
`model = "claudeAgent/…"`): поменять id. Внимание: t3code для незнакомого
|
|
||||||
ему слага теряет опции `effort`/`contextWindow` (см. ниже) — id должен
|
|
||||||
быть в каталоге t3code, т.е. t3code-приложение должно быть обновлено.
|
|
||||||
4. Деплой: `make deploy` (push `main:stable`; образ гейтвея собирается из
|
|
||||||
`beaver-gateway#main` при деплое, поэтому шаг 1 — раньше).
|
|
||||||
5. **t3code на машинах**: `claude --version` ≥ минимума модели (обычно уже
|
|
||||||
сам обновился); само приложение t3code (alpha) обновляется nightly и
|
|
||||||
привозит новую модель в каталог.
|
|
||||||
|
|
||||||
Проверить, какие id вообще существуют:
|
|
||||||
|
|
||||||
```bash
|
|
||||||
TOK=$(security find-generic-password -s "Claude Code-credentials" -w | jq -r .claudeAiOauth.accessToken)
|
|
||||||
curl -s https://api.anthropic.com/v1/models -H "Authorization: Bearer $TOK" \
|
|
||||||
-H "anthropic-version: 2023-06-01" -H "anthropic-beta: oauth-2025-04-20" | jq -r '.data[].id'
|
|
||||||
```
|
|
||||||
|
|
||||||
## Алиасы `fable` / `opus` / `sonnet`
|
|
||||||
|
|
||||||
CLI принимает голые алиасы и резолвит их в **новейшую модель, которую знает
|
|
||||||
установленный CLI** (проверено: CLI 2.1.248 резолвит `fable` →
|
|
||||||
`claude-fable-5`, не 5.1). Свойства:
|
|
||||||
|
|
||||||
- никогда не дают 400 «слишком старый клиент» — деградируют до старой модели;
|
|
||||||
- «всегда новейшая» они дают только при свежем CLI, так что шаг 1 чеклиста
|
|
||||||
они не отменяют, но убирают шаги 2–3 (id в конфигах менять не надо);
|
|
||||||
- **в t3code** кастомный слаг (`fable`, `opus`, как и любой id вне каталога)
|
|
||||||
имеет пустые capabilities → t3 не передаёт `--effort` и суффикс `[1m]`
|
|
||||||
(`ClaudeAdapter.ts`: `resolveClaudeEffort(caps,…)` → undefined). Модель
|
|
||||||
работает, но на дефолтном effort и дефолтном окне CLI.
|
|
||||||
|
|
||||||
Рекомендация: в `config.py` и `t3code.toml` можно перейти на алиасы, если
|
|
||||||
устраивает «новейшее из того, что знает CLI»; explicit id — когда нужен
|
|
||||||
точный контроль (и тогда обязательно проверять минимум версии CLI).
|
|
||||||
@@ -2,7 +2,7 @@
|
|||||||
|
|
||||||
My real beaver-gateway setup (paired with the protocol from beaver.kotikot.com).
|
My real beaver-gateway setup (paired with the protocol from beaver.kotikot.com).
|
||||||
|
|
||||||
Uses Claude Code and Raycast subscriptions for agents, has some MCPs set up. You can use this as-is or modify to match your needs (config.py).
|
Uses Claude Code and Raycast subscriptions for agents, has some MCPs set up. You can use this as-is or modify to match your needs: `config.py` assembles the gateway from the `beaver_agent/` package.
|
||||||
|
|
||||||
## Requirements
|
## Requirements
|
||||||
|
|
||||||
@@ -60,12 +60,14 @@ regular Claude Code; rotate it by running `claude setup-token` again.
|
|||||||
|
|
||||||
Claude agents run on the Claude Agent SDK. Prompts are granules under
|
Claude agents run on the Claude Agent SDK. Prompts are granules under
|
||||||
`мета/бобер/промпты/` in the vault - plain markdown, the `<tag>` wrappers
|
`мета/бобер/промпты/` in the vault - plain markdown, the `<tag>` wrappers
|
||||||
and the order live in `config.py`; skills are the folders under
|
and the order live in `beaver_agent/prompts.py`; skills are the folders under
|
||||||
`мета/бобер/скиллы/` (each becomes a plugin, keep `name:` in SKILL.md
|
`мета/бобер/скиллы/` (each becomes a plugin, keep `name:` in SKILL.md
|
||||||
latin). The vault must be synced before the gateway can start. The model process runs as `beaver-runner`
|
latin), which window loads which set is `beaver_agent/skills.py`. The vault
|
||||||
with a whitelisted environment; the vault is mounted read-write (it is the
|
must be synced before the gateway can start. The model process runs as
|
||||||
dispatcher's home), and `policy.py` keeps it out of `.obsidian`, `мета`
|
`beaver-runner` with a whitelisted environment; the vault is mounted
|
||||||
outside `мета/бобер`, and from deleting anything but its own files.
|
read-write (it is the dispatcher's home), and `beaver_agent/policy.py` keeps
|
||||||
|
it out of `.obsidian`, `мета` outside `мета/бобер`, and from deleting
|
||||||
|
anything but its own files.
|
||||||
|
|
||||||
### 4. Mint a token
|
### 4. Mint a token
|
||||||
|
|
||||||
@@ -131,30 +133,55 @@ Needs `KOMODO_URL` / `KOMODO_KEY` / `KOMODO_SECRET` in `.env` (and
|
|||||||
inject. Jobs, the queue and the subscription window are on the admin **Jobs** page.
|
inject. Jobs, the queue and the subscription window are on the admin **Jobs** page.
|
||||||
|
|
||||||
The same key backs the `komodo` tool the dispatcher and deep chats get
|
The same key backs the `komodo` tool the dispatcher and deep chats get
|
||||||
(`mcps/komodo.py`, архитектура §4.4): one MCP tool with an enumerated
|
(`beaver_agent/hands/komodo.py`): one MCP tool with an enumerated
|
||||||
`action` - `status`, `stacks`, `containers`, `logs`, `search_logs`, `updates`,
|
`action` - `status`, `stacks`, `containers`, `logs`, `search_logs`, `updates`,
|
||||||
`update`, `deploy`, `restart`, `exec`. `exec` is `docker exec` into any
|
`update`, `deploy`, `restart`, `exec`. `exec` is `docker exec` into any
|
||||||
container of the fleet except infrastructure ones (`exec_deny`, periphery by
|
container of the fleet except infrastructure ones (`exec_deny`, periphery by
|
||||||
default); prune, destroy and a host terminal do not exist in the enumeration,
|
default); prune, destroy and a host terminal do not exist in the enumeration,
|
||||||
so they cannot be asked for. The key never reaches the model process.
|
so they cannot be asked for. The key never reaches the model process.
|
||||||
|
|
||||||
## Policy (`policy.py`)
|
## Layout: `config.py` + `beaver_agent/`
|
||||||
|
|
||||||
`bypassPermissions` everywhere; the boundary is the read-only vault mount
|
`config.py` is a page: it imports the pieces and calls `Gateway(...)`. The
|
||||||
plus `PreToolUse` rules declared per agent (`ClaudeAgent.policy`, архитектура
|
pieces live in `beaver_agent/`, one concern per module:
|
||||||
§3.7): writes only under `мета/бобер/`, new files only under `💬 чаты/`,
|
|
||||||
`rm`/`mv`/`cp`/`tee`/`sed -i`/redirects into the vault outside those zones
|
|
||||||
are refused with a reason, and `mcp__firefly__store_*`/`update_*` need the
|
|
||||||
`firefly` skill opened first in the same session. Every tool call lands in
|
|
||||||
the admin **Audit** page as `tool_call`. Run `make check` - it includes the
|
|
||||||
policy and komodo tests.
|
|
||||||
|
|
||||||
## Editing config.py on the mac
|
- `vault.py` - paths and the agent's zone, `TZ`
|
||||||
|
- `prompts.py` - how each role's system prompt is assembled from granules
|
||||||
|
- `skills.py` - which window loads which skill set
|
||||||
|
- `policy.py` - `PreToolUse` rules and vault zones (`ZONES`, `VAULT_POLICY`)
|
||||||
|
- `hands/` - every MCP the setup has and which agent gets which one; the
|
||||||
|
connectors (komodo, home assistant, vibegram) live here too
|
||||||
|
- `agents.py` - the roles (dispatcher, deep, distiller, curator, triage,
|
||||||
|
raycast) and the list of instances with models and effort
|
||||||
|
- `frontends.py` - the windows: Telegram for master and branches, vault
|
||||||
|
files for deep chats, `/api`, `/admin`, `/anthropic`, `/mcp`
|
||||||
|
- `texts.py` - everything the gateway says to the model or to the user,
|
||||||
|
in Russian (the gateway's own defaults are English)
|
||||||
|
- `memory/` - the recall block under a message, the reply log, handouts
|
||||||
|
and seeds, the vault watcher, the curator briefing
|
||||||
|
- `jobs/` - crons and webhooks, one file per job
|
||||||
|
|
||||||
|
## Policy (`beaver_agent/policy.py`)
|
||||||
|
|
||||||
|
`bypassPermissions` everywhere; the boundary is `PreToolUse` rules declared
|
||||||
|
per agent (`ClaudeAgent.policy`): the vault is the dispatcher's home, so it
|
||||||
|
may create, edit and move anywhere except `.obsidian` and `мета` outside
|
||||||
|
`мета/бобер`, and may delete only its own files; the distiller and curator
|
||||||
|
stay in `мета/бобер`; `rm`/`mv`/`cp`/`tee`/`sed -i`/redirects into the vault
|
||||||
|
outside those zones are refused with a reason; `mcp__firefly__store_*` and
|
||||||
|
`update_*` need the `firefly` skill opened first in the same session, and the
|
||||||
|
master cannot open the `vault/` skills at all (a branch can). Every tool call
|
||||||
|
lands in the admin **Audit** page as `tool_call`. Run `make check`.
|
||||||
|
|
||||||
|
## Editing the setup on the mac
|
||||||
|
|
||||||
`make sync` installs `../beaver-gateway` editable (extra `local`), so
|
`make sync` installs `../beaver-gateway` editable (extra `local`), so
|
||||||
imports in `config.py` resolve against the checkout you are working on;
|
imports resolve against the checkout you are working on; `uv sync --extra
|
||||||
`uv sync --extra prod` pulls the gateway from git instead. Neither is used
|
prod` pulls the gateway from git instead. Neither is used by the image - the
|
||||||
by the image - the container builds the gateway itself.
|
container builds the gateway itself. To assemble the config against your
|
||||||
|
local vault: `BEAVER_VAULT=~/path/to/vault uv run python -c
|
||||||
|
'from beaver_gateway import config; from pathlib import Path;
|
||||||
|
config.load(Path("config.py"))'` with a `.env` next to it.
|
||||||
|
|
||||||
## Useful commands
|
## Useful commands
|
||||||
|
|
||||||
@@ -175,13 +202,13 @@ docker compose logs -f obsidian-headless
|
|||||||
# pull the latest beaver-gateway (when GATEWAY_REF=main)
|
# pull the latest beaver-gateway (when GATEWAY_REF=main)
|
||||||
docker compose build gateway && docker compose up -d gateway
|
docker compose build gateway && docker compose up -d gateway
|
||||||
|
|
||||||
# reload config.py without a full rebuild
|
# reload config.py / beaver_agent without a full rebuild
|
||||||
docker compose restart gateway
|
docker compose restart gateway
|
||||||
```
|
```
|
||||||
|
|
||||||
## Gotchas
|
## Gotchas
|
||||||
|
|
||||||
- **claude in the container doesn't see the vault** - `cwd=VAULT` in `config.py` resolves to `/vault` *inside* the container, not on the host. Don't change it.
|
- **claude in the container doesn't see the vault** - `VAULT` in `beaver_agent/vault.py` is `/vault` *inside* the container, not on the host. Don't change it; `BEAVER_VAULT` is for assembling the config on the mac only.
|
||||||
- **a claude turn fails immediately** - check `docker logs beaver-gateway` for the `claude[<agent>]:` stderr lines. Usually auth: redo step 3.
|
- **a claude turn fails immediately** - check `docker logs beaver-gateway` for the `claude[<agent>]:` stderr lines. Usually auth: redo step 3.
|
||||||
- **gateway restarts in a loop right after first `up`** - `config.py` reads prompt granules from `/vault/мета/бобер/промпты`; until Obsidian Sync has pulled the vault they are missing. Finish step 2, it settles.
|
- **gateway restarts in a loop right after first `up`** - `beaver_agent/prompts.py` reads prompt granules from `/vault/мета/бобер/промпты`; until Obsidian Sync has pulled the vault they are missing. Finish step 2, it settles.
|
||||||
- **the model cannot write into `мета/бобер`** - the entrypoint grants `beaver-runner` an ACL on the two rw sub-mounts once at start; files that Sync creates later inherit it through the default ACL. If `setfacl` is unsupported on the volume it falls back to `chown`, and then files Sync writes afterwards as root stay read-only for the model until the next restart.
|
- **the model cannot write into `мета/бобер`** - the entrypoint grants `beaver-runner` an ACL on the two rw sub-mounts once at start; files that Sync creates later inherit it through the default ACL. If `setfacl` is unsupported on the volume it falls back to `chown`, and then files Sync writes afterwards as root stay read-only for the model until the next restart.
|
||||||
|
|||||||
Reference in New Issue
Block a user