agim-cli 1.2.44 → 1.2.48

package/CHANGELOG.md

@@ -4,6 +4,301 @@ All notable changes to this project will be documented in this file.
 
 ## [Unreleased]
 
+## [1.2.48] - 2026-05-24
+
+### Theme: native+ — production-ready in-process LLM agent
+
+Wraps up the post-v1.3.0 / v1.2.47 follow-ups into one drop. Native is
+now the first-class peer of `claude-code` / `opencode` / `codex` for
+end-to-end IM ↔ multi-agent work: it knows its identity, sees every
+skill the box has, talks to vision models, recovers from provider
+hiccups, summarises stale chat, and never silently dies on a long
+multi-hop research turn.
+
+### Added
+
+- **Operator role definition** — `<native cwd>/AGENTS.md` is now read
+  on every turn and prepended to the system prompt as
+  `[operator role definition]`. Edit the seed at
+  `~/.agim-workspaces/native/AGENTS.md` to change persona / tone /
+  house rules; no restart required. Env override
+  `IMHUB_NATIVE_AGENT_ROLE_FILE=...` for non-default locations.
+  Prompt-injection guard runs over the content + audit-logs hits.
+- **Skills multi-root scan** — `IMHUB_SKILLS_MODE=auto` (default)
+  merges `~/.agim/skills/` (write target) with `~/.claude/skills/`,
+  `~/.config/opencode/skills/`, and `~/.codex/skills/` so a fresh
+  install sees every skill the user already has from CLI agents.
+  Workspace skills shadow same-named entries from inherited roots.
+  Modes: `auto | agim-only | custom` + `IMHUB_SKILLS_ROOTS=...`
+  / `IMHUB_SKILLS_EXTRA_ROOTS=...`.
+- **Auto-compact long histories** — when a native turn's accumulated
+  history crosses `IMHUB_NATIVE_COMPACT_TRIGGER_CHARS` (default
+  60_000), a cheap LLM call summarises the older slice and the loop
+  keeps only the last `IMHUB_NATIVE_COMPACT_KEEP_TURNS` (default 6)
+  user/assistant pairs verbatim. Tool round-trips kept intact (the
+  fix that landed in CR — never start the tail on `role:'tool'`).
+- **Provider fallback chain** — native walks the configured role
+  chain (e.g. `native-chat` → `cheap` → `evaluator`) on transient
+  provider errors (5xx / timeout / ECONNRESET) so a flaky OpenAI
+  endpoint doesn't kill the whole turn. Logged as
+  `native.fallback.next`.
+- **Multimodal pass-through (OpenAI-compat)** — messenger image
+  attachments (`[图片附件:...]` markers) are encoded as `image_url`
+  content blocks when the provider's `vision: true` (or its model
+  name matches a known vision family). Anthropic provider remains
+  text-only for now. Hard cap 20MB per image. Path-safety:
+  attachments must resolve under `IMHUB_MEDIA_ROOT`.
+- **Tool-call heartbeat** — tool dispatches that take longer than
+  `IMHUB_NATIVE_HEARTBEAT_MS` (default 6000) push a "🔧 调用工具 X
+  中…" so users know the bot is still working.
+- **Turn-level heartbeat** — every
+  `IMHUB_NATIVE_TURN_HEARTBEAT_MS` (default 180_000 = 3 min) since
+  turn start, push "⏳ 还在处理（已 Nm）…". Survives long
+  multi-hop research turns where individual tool calls finish but the
+  outer LLM is reasoning.
+- **Cross-channel operator alert** — when outbox delivery fails ≥3
+  times on any thread, push a "⚠️ outbox 告警" to
+  `IMHUB_OPERATOR_ALERT_PLATFORM` / `IMHUB_OPERATOR_ALERT_THREAD_ID`
+  (throttled per dead-thread, 10-min cooldown). Self-loop protection.
+- **agim built-in MCP tools default-allowed** — the 9
+  `mcp__imhub__*` tools (read_skill / list_skills / save_memo /
+  search_memos / update_memo / delete_memo / push_message /
+  ask_user / call_agent) are now in native's default allow-list,
+  matching what claude-code already saw via the MCP sidecar. Override
+  via `IMHUB_NATIVE_AGENT_DENYLIST=...`.
+- **Per-agent native workspace** — `resolveAgentCwd('native', opts)`
+  pins IM turns to `~/.agim-workspaces/native/` (was `/` under
+  systemd). AGENTS.md seeded with role-definition cues.
+- **Web admin: skills merged into Settings → 技能** — top-level
+  `/skills` nav entry retired. New layout has 2 subtabs (Installed +
+  skillhub 热门). Old URLs redirect.
+
+### Fixed
+
+- **access-token paths now lazy** (v1.2.47 carry-over) — `tokensFile()`
+  / `webBootstrapTokenFile()` re-read `process.env.AGIM_HOME` per
+  call so tests that swap home directories actually redirect writes.
+  Exports added: `getTokensFilePath()`, `getWebBootstrapTokenFile()`,
+  `__resetBootstrapHolderForTesting()`.
+- **agent-loop default 5-min timeout no longer cuts native turns**
+  — native now passes `timeoutMs: 28 * 60 * 1000` to `runAgentLoop`,
+  matching the IM 30-min hard ceiling. Multi-hop `call_agent` chains
+  used to abort mid-flight on the 6th minute. Override via
+  `IMHUB_NATIVE_AGENT_TIMEOUT_MS`.
+- **wechat-ilink dead-session auto-recovery** — `sendMessage` returning
+  `-2` (`INVALID_PARAMS`) now invalidates the cached + persisted
+  context-token for that user, so the next inbound message refreshes
+  it and the outbox retry succeeds. Was getting stuck in retry
+  forever previously.
+- **render-router stray envelope tags** — captured `<im-summary>` /
+  `<full-md>` content no longer leaks closing tags into the IM
+  bubble (regression when LLM emitted multiple `<im-summary>`
+  blocks back-to-back).
+- **Auto-compact tool-orphan bug** (caught by CR) — the slice index
+  picker now strictly aligns to `role:'user'` so the recent tail
+  never starts on a `role:'tool'` whose matching `assistant.toolCalls`
+  was just compacted away (would otherwise hit HTTP 400 on
+  OpenAI / DeepSeek).
+- **parsePromptMedia path-safety hardened** — `path.resolve` +
+  separator-aware prefix check now rejects sibling-directory leaks
+  (e.g. `/root/.agim/media-evil/...`) and `..` traversal.
+- **ESM `require` → static `import`** — `parsePromptMedia` and
+  `mediaToDataUrl` use top-of-file `import` (was crashing at
+  runtime with `require is not defined`).
+
+### Migration
+
+- Operators who customised `~/.agim-workspaces/native/AGENTS.md` keep
+  their file unchanged (`ensureAgentWorkspace` never overwrites).
+- Default skill mode is `auto` — agim now reads `~/.claude/skills/`
+  etc. Set `IMHUB_SKILLS_MODE=agim-only` to opt out of inheritance.
+- The 9 `mcp__imhub__*` tools are auto-allowed under native's
+  default policy. Set `IMHUB_NATIVE_AGENT_DENYLIST=mcp__imhub__call_agent,...`
+  to block individual ones.
+- Auto-compact will make one cheap-role LLM call per compaction.
+  Disable with `IMHUB_NATIVE_COMPACT_TRIGGER_CHARS=0`.
+
+## [1.3.0] - 2026-05-23
+
+### Theme: nanobot-vs-agim comparison rollout — native LLM stack + P0–P3
+
+Single-version landing for the multi-stage work that adopted the most
+valuable structural features from `HKUDS/nanobot` **without changing
+agim's product identity**. agim stays a messenger ↔ agent bridge; the
+new `native` agent is a peer `AgentAdapter` alongside the CLI agents.
+See [`docs/nanobot-comparison-rollout.md`](docs/nanobot-comparison-rollout.md)
+for the PR-by-PR landing order and
+[`docs/architecture-bridge-and-native.md`](docs/architecture-bridge-and-native.md)
+for the architectural stance + anti-patterns.
+
+### Added — native LLM stack (Stage 1 + 2)
+
+- **`native` AgentAdapter** — in-process LLM agent loop, no subprocess.
+  Registered alongside `claude-code` / `codex` / `opencode` /
+  `antigravity` and reachable via `/cc native` / `/na` / `/llm`.
+  `kind='in-process'` UI tag only; router / approval / A2A / audit
+  code stays kind-agnostic. PR #65 + #73.
+- **LLM provider stack** — `openai-compat-provider` (DeepSeek / Kimi /
+  Qwen / Ollama / vLLM / OpenAI), `anthropic-provider`, common
+  `provider-base` with retries + abort + token usage. PR #59 + #62.
+- **Multi-iteration agent loop** — `agent-loop.ts` runs `chat → tool
+  calls → tool results → chat …` until done; max-turns guard; per-turn
+  approval gate; reasoning_content round-trip for DeepSeek tool-call
+  turns (`ed36b39`). PR #64.
+- **External MCP server attach** — `mcp-client.ts` + `mcp-registry.ts`
+  + `mcp-config.json` connect stdio / HTTP / SSE MCP servers and
+  surface their tools as `mcp_<server>_<tool>` to the native
+  dispatcher. PR #63.
+- **Policy approval gate** — `policy-approval-gate.ts` with
+  `mode=allow-list | deny-list | always-allow | always-deny`,
+  per-tool `autoAllow` count, audit-logged decisions. PR #64.
+- **`llmBackends` + `llmRoles` + `~/.agim/llm-secrets.json`** — config
+  schema for backend definitions, role bindings (`cheap` / `evaluator`
+  / `native-chat`), and secret storage with 0o600 perms + redacted
+  list endpoints. PR #59 + #70.
+- **`tryIntrospectChain` fallback helper** — internal callers
+  (`memory-distill`, `memory-consolidate`, `intent-llm`) try roles in
+  order and fall back gracefully when a backend is unavailable. PR #71.
+- **`loadFromConfig` boot wiring** — `src/cli.ts` now calls
+  `loadLlmFromConfig` + `loadMcpFromConfig` + `connectAll` after
+  built-in plugin load so `native.isAvailable()` returns true on first
+  boot when a backend is configured. `11cc0f6`.
+
+### Added — P0 items riding on the stack
+
+- **`/heartbeat` per-thread bindings + 3-phase scheduler (P0 #1)** —
+  bind a thread to a tick interval; each tick runs DECIDE (cheap LLM)
+  → EXECUTE (the agent does the work) → EVALUATE (notification gate
+  decides whether to ping the user). `/tasks/heartbeat` cross-thread
+  admin. PR #61.
+- **`/goal` sustained-objective tracker (P0 #2)** — one active goal
+  per IM thread, auto-injected into every prompt for cross-turn
+  context. Lifecycle `set / body / progress / pause / resume /
+  complete / cancel / list`. `/tasks/goals` cross-thread admin. See
+  [`docs/goals.md`](docs/goals.md). PR #66.
+- **agim Skills engine (Stage 3 + P0 #3)** — discover `SKILL.md`
+  files in `~/.agim/skills/` + bundled `<dist>/skills/builtin/`; 3-tier
+  progressive disclosure (name+desc auto-injected; body via
+  `mcp__imhub__read_skill`; scripts/refs via normal file read);
+  `/skill list|show|refresh` + `/settings/agim-skills` web editor. See
+  [`docs/skills.md`](docs/skills.md). PR #68 + #76.
+- **Inline row-age annotations on memory facts (P0 #4)** — every fact
+  / persona row gets a `(2d ago)` style annotation in the injected
+  context so the agent has temporal context for free. `format-age.ts`
+  utility + `persona-age.test.ts`. PR #69.
+- **`mcp__imhub__ask_user` structured choice (P0 #5)** — agent presents
+  N choices (2-9) with optional timeout; user replies via button or
+  text; result returns deterministically as a typed choice index +
+  free-text. `/tasks/asks` pending-queue admin. See
+  [`docs/ask-user.md`](docs/ask-user.md). PR #67.
+- **Notification evaluator for agent-initiated push (P0 #9)** —
+  `mcp__imhub__push_message` and `POST /api/notify` go through an
+  LLM-graded "should we deliver this?" filter before hitting the IM
+  thread. Per-thread rate limit; verdict logged to audit. PR #60.
+
+### Added — config UX (Stage 4 + 6) and CLI symmetry (P2/P3)
+
+- **Web settings pages** —
+  - `/settings/llm` — list / add / remove backends, role bindings,
+    secret slots, per-backend **Ping** via
+    `/api/llm/backends/:name/ping`.
+  - `/settings/mcp` — list / add / remove external MCP servers,
+    per-server **Health** via `/api/mcp/servers/:name/health`.
+  - `/settings/native-agent` — focused form for native-agent /
+    notification-evaluator / heartbeat env knobs.
+  - `/settings/agim-skills` — SKILL.md editor.
+  PR #74 + #76.
+- **Web task views** — `/tasks/heartbeat`, `/tasks/goals`,
+  `/tasks/asks` cross-thread admin views with auto-refresh and
+  per-row actions. REST `/api/heartbeat/bindings`, `/api/goals`,
+  `/api/asks/pending`. PR #75.
+- **CLI wizards**:
+  - `agim setup llm {add,list,remove,role,secret,native-enable}` —
+    full LLM backend + role + secret management; `native-enable`
+    one-shot enables the native agent if a backend exists. PR #70 + #77.
+  - `agim setup mcp {add,list,remove,health}` — MCP server CRUD +
+    probe. Restores CLI/Web symmetry. PR #78.
+  - `agim diag {llm,mcp}` — ping every backend / connect-and-list-tools
+    per MCP server with single-screen status output. PR #77.
+- **Config wizard now offers `native`** — `agim` first-run wizard
+  detects `llmBackends` and offers `native` as a checkbox agent
+  alongside `claude-code` etc. `03a04f5`.
+- **`gen-endpoints --check` drift guard** — `tools/gen-endpoints.mjs`
+  added a `--check` mode that fails CI when generated endpoint
+  constants drift from the SPA hand-written exports. PR #79.
+
+### Added — Stage 5 router
+
+- **`/router compare agent:<a> agent:<b> "<prompt>"`** — A/B evaluator
+  runs the same prompt against two backends side-by-side (any combo of
+  CLI and native), returns a single card with latency + tokens + raw
+  outputs so operators can pick a winner before changing
+  `defaultAgent`. PR #71.
+
+### Added — i18n
+
+- **Full bilingual (en / zh) coverage** for every new v1.3.0 surface:
+  - Web SPA — `/settings/{llm,mcp,native-agent,agim-skills}` +
+    `/tasks/{heartbeat,goals,asks}` use `useTranslation` with the new
+    `settings.json` / `tasks.json` locale namespaces.
+  - CLI — `src/cli-ui/i18n.ts` adds native/MCP startup logs, agent
+    install hints (`formatAgentInstallHint(lang, missing)`),
+    `agim token create/list/revoke/bootstrap` output, config wizard
+    `native` agent label, and `resolveCliLang()` for env-derived
+    lang. `0eef14e`.
+- **`agim token bootstrap`** prints the one-shot
+  `~/.agim/admin-bootstrap-token` to the operator on the host. PR #76
+  follow-up + `6490ce5`.
+
+### Architecture
+
+- **`AgentAdapter.kind: 'cli' | 'in-process' | 'remote'`** — UI tag
+  only; router / approval bus / A2A / audit code does NOT branch on
+  it. New `describe?()` method renders the role/model line in
+  `/agents`. PR #73.
+- **`acp-token` separated from `web-token`** — the ACP bearer now
+  lives in `~/.agim/acp-token` (legacy fallback `~/.agim/web-token`),
+  giving operators an independent rotation surface for remote-agent
+  inbound traffic.
+
+### Docs
+
+- `docs/architecture-bridge-and-native.md` (new) — peer
+  `AgentAdapter` design, the six things every backend gets for free
+  (routing / A2A / approval / audit / `/router compare` / fallback
+  chains), anti-patterns.
+- `docs/nanobot-comparison-rollout.md` (new) — single-page reference
+  for the multi-stage rollout, recommended PR landing order, end-state
+  capability matrix, deferred items.
+- `docs/llm-backends.md` (new) — native LLM backend configuration,
+  provider matrix, secret handling, role bindings, troubleshooting.
+- `docs/goals.md`, `docs/ask-user.md`, `docs/skills.md` (new) — full
+  primitive references.
+
+### Fixed
+
+- **`native.isAvailable()` returns true on first boot** — `cli.ts`
+  was calling `loadBuiltInPlugins()` but never `loadLlmFromConfig`,
+  so the in-process LLM registry stayed empty and the native agent
+  reported unavailable until something else lazily loaded it.
+  `11cc0f6`.
+- **DeepSeek tool-call turn `reasoning_content` round-trip** —
+  preserve the assistant `reasoning_content` field across tool-call
+  turns so multi-iteration loops don't drop chain-of-thought when the
+  model returns it alongside a tool call. `ed36b39`.
+- **SPA registers v1.3.0 routes** — `/settings/{llm,mcp,native-agent,
+  agim-skills}` and `/tasks/{heartbeat,goals,asks}` registered in the
+  v2 SPA router. `c97b023` + `d29ee47`.
+
+### Compatibility
+
+- Native agent is **opt-in by default**; without `llmBackends` the
+  bridge behaviour is identical to v1.2.x.
+- All v1.2.x configs continue to work without edit; the new
+  `llmBackends` / `llmRoles` keys are additive.
+- `~/.agim/web-token` is still honored as a fallback for the ACP
+  bearer if `~/.agim/acp-token` does not exist.
+
 ## [1.2.32] - 2026-05-22
 
 ### R15 — final R14 follow-ups

package/README.md

@@ -18,6 +18,14 @@
 ## Highlights
 
 - **3 messengers + email, 3+ agents** — WeChat (image / file / voice), Feishu, DingTalk (image / voice with server-side ASR), Email (SMTP); Claude Code, Codex, OpenCode, plus any ACP endpoint
+- **Native LLM stack (v1.3.0+, hardened in v1.2.48)** — built-in in-process agent loop (`native` adapter) that talks directly to any OpenAI- / Anthropic-compatible API (DeepSeek, Kimi, GLM, Qwen, Ollama, vLLM, OpenAI, Anthropic). v1.2.48 makes native production-ready: **operator role definition** via `~/.agim-workspaces/native/AGENTS.md` (hot-reloaded, no restart), **multimodal vision** (image attachments → openai-compat `image_url` blocks), **auto-compact** long histories (>60k chars summarised by cheap role), **provider fallback chain** on transient 5xx/timeout, **turn-level heartbeat** every 3 min so multi-hop research turns never look dead, **28-min agent-loop timeout** (was 5-min default, killed long `call_agent` chains). Use it as a cheap router for `memory-distill` / `memory-consolidate` / `intent-llm`, as a local-first chat agent (`/na`), or as a peer in A2A. See [`docs/llm-backends.md`](docs/llm-backends.md) + [`docs/architecture-bridge-and-native.md`](docs/architecture-bridge-and-native.md).
+- **Skills multi-root scan (v1.2.48+)** — `IMHUB_SKILLS_MODE=auto` (default) merges `~/.agim/skills/` (write target) with `~/.claude/skills/`, `~/.config/opencode/skills/`, `~/.codex/skills/` so a fresh install sees every skill you already have from CLI agents. Workspace skills shadow same-named inherited entries. Settings → 技能 has 2 subtabs (Installed editor + skillhub 热门); top-level `/skills` nav merged into Settings.
+- **Cross-channel outbox alert (v1.2.48+)** — when a thread's outbox is dead (e.g. wechat session expired), agim auto-invalidates the stale context-token + pushes a "⚠️ outbox 告警" to your configured backup channel (`IMHUB_OPERATOR_ALERT_PLATFORM` / `IMHUB_OPERATOR_ALERT_THREAD_ID`) so you find out before the user does.
+- **Per-thread sustained goals (v1.3.0+ · `/goal`)** — one active multi-turn objective per IM thread, auto-injected into every prompt so the agent stays on-mission across many user turns. Lifecycle: `set / body / progress / pause / resume / complete / cancel / list`. Cross-thread admin at `/tasks/goals`. See [`docs/goals.md`](docs/goals.md).
+- **Heartbeat scheduler (v1.3.0+ · `/heartbeat`)** — bind a thread to a periodic "anything to do right now?" tick. Each tick runs DECIDE (cheap LLM) → EXECUTE (the agent does the work) → EVALUATE (notification gate decides whether to ping the user). Cross-thread admin at `/tasks/heartbeat`.
+- **Structured agent → user choice (v1.3.0+ · `mcp__imhub__ask_user`)** — agents present an N-way multiple-choice question and reliably get a typed answer back (button + text reply, configurable timeout). Avoids the "free-form chat" ambiguity for disambiguation. Pending-queue admin at `/tasks/asks`. See [`docs/ask-user.md`](docs/ask-user.md).
+- **agim Skills engine (v1.3.0+ · `/skill`)** — drop a `SKILL.md` into `~/.agim/skills/<name>/`, every CLI agent + native sees it. 3-tier progressive disclosure (name+desc always-on, body on-demand via `mcp__imhub__read_skill`, scripts/refs via normal file-read). Web editor at `/settings/agim-skills`. See [`docs/skills.md`](docs/skills.md).
+- **`/router compare` A/B evaluator (v1.3.0+)** — run the same prompt against two backends side-by-side (any combo of CLI and native), get latency + tokens + outputs in one card so you can pick a winner before changing `defaultAgent`.
 - **Compliance-first defaults (v1.2.23+)** — global IM adapters (Discord, Telegram) and the remote-ACP-agent surface are **hidden** by default; no installer prompt or web-admin card exposes them. Mainland-China deployments should leave them disabled. For cross-border setups, opt-in via `IMHUB_ENABLE_GLOBAL_IM=1` / `IMHUB_ENABLE_REMOTE_AGENT=1`. See [`docs/security-hardening.md`](docs/security-hardening.md) for the full security control catalogue (~30 env vars / sensitive-path denylist / outbound secret scrubber / audit-events table / admin role on tokens / R14 systemctl-aware lifecycle).
 - **Agent-to-Agent (A2A)** — agents call other agents inline via `mcp__imhub__call_agent`. Just say *"用 codex 帮我跑 git status"* / *"ask opencode to run the tests"* — the active agent hands off, waits, integrates the reply. Guardrails: depth limit (default 3), self-call ban, workspace whitelist, per-user budget. Observability: `/a2a stats | recent | tree <id>`. Disable with `IMHUB_A2A_ENABLED=0`.
 - **A2A shared artifacts (v1.1.3+)** — caller drops files into the callee's workspace via `inputs[]` (`fromAbsolutePath` / `fromCallerOutput` / inline `content`), and callee writes products to `~/.agim/artifacts/<jobId>/_agim-output/`. Returned `outputs[]` lets the caller read files on demand. Hard-link first, copy fallback. Size caps configurable.
@@ -90,15 +98,18 @@ units that call `agim` continue to work without any extra setup.
 | Category | Details |
 |----------|---------|
 | **Messengers** | WeChat (iLink — image / file / voice / video), Feishu (WebSocket), DingTalk (Stream mode — image / voice with built-in ASR), Email (SMTP, push-only) |
-| **Agents** | Claude Code, Codex, OpenCode — all via shared `AgentBase`; any HTTP agent via ACP |
+| **Agents** | Claude Code, Codex, OpenCode, Antigravity (all CLI via shared `AgentBase`); **`native`** in-process LLM loop; any HTTP agent via ACP — all share one `AgentAdapter` interface |
+| **Native LLM** | OpenAI- + Anthropic-compatible providers, external MCP server attach, multi-iteration agent loop with policy approval gate, role bindings (`cheap` / `evaluator` / `native-chat`) |
 | **Reminders** | `/remind` slash, LLM intent detection, LLM-polished delivery, MCP tools for agents (claude-code + opencode), web `/reminders` UI |
-| **Web UI** | Chat with streaming, three-state theme (light / dark / system), bilingual (EN / 中文), SSE dashboard |
-| **Tool Approval** | Human-in-the-loop over IM + in-page cards; MCP sidecar for Claude |
+| **Goals + Heartbeat + Ask** | `/goal` per-thread objective, `/heartbeat` periodic DECIDE/EXECUTE/EVALUATE tick, `mcp__imhub__ask_user` structured multi-choice with `/tasks/{goals,heartbeat,asks}` cross-thread admin |
+| **Skills** | `~/.agim/skills/<name>/SKILL.md` discovered by every agent (CLI + native); 3-tier progressive disclosure; web editor at `/settings/agim-skills` |
+| **Web UI** | Chat with streaming, three-state theme (light / dark / system), bilingual (EN / 中文 — every new v1.3.0 page included), SSE dashboard, settings tabs for LLM / MCP / Native Agent / Skills |
+| **Tool Approval** | Human-in-the-loop over IM + in-page cards; MCP sidecar for Claude; allow-list policy gate for native |
 | **Jobs** | Persistent SQLite job board + cron scheduler; batch ops; background task reader |
-| **Routing** | Intent classifier, circuit breaker, rate limiter, sticky sessions, LLM judge fallback |
+| **Routing** | Intent classifier, circuit breaker, rate limiter, sticky sessions, LLM judge fallback; `/router compare` A/B evaluator |
 | **Workspaces** | Multi-tenant; agent whitelist + rate limits; command-level ACL |
 | **Observability** | pino structured logs, traceId, Prometheus `/api/metrics`, SQLite audit log |
-| **ACP** | Client (connect to remote agents) + Server (Agim itself as an ACP agent) |
+| **ACP** | Client (connect to remote agents) + Server (Agim itself as an ACP agent) with token-gated `~/.agim/acp-token` (legacy fallback `~/.agim/web-token`) |
 
 ## CLI Commands
 
@@ -117,6 +128,21 @@ agim config claude         # Configure Claude Code
 agim config agent          # Connect a custom ACP agent
 agim agents                # List available agents
 agim messengers            # List available messengers
+
+# v1.3.0+ — native LLM stack & diagnostics
+agim setup llm             # Wizard: add/list/remove LLM backends, bind roles, set secrets
+agim setup llm native-enable  # One-shot: enable `native` agent if a backend is configured
+agim setup mcp             # Wizard: add/list/remove external MCP servers (stdio / HTTP / SSE)
+agim setup mcp health      # Probe every configured MCP server
+agim diag llm              # Ping every backend + show which role binds to what
+agim diag mcp              # Connect-and-list-tools per MCP server
+
+# v1.1.10+ — Web/ACP token management
+agim token create [name]   # Mint a new user/admin token for /api/* + /acp/*
+agim token list            # List tokens (last 4 chars only)
+agim token revoke <id>     # Revoke a token by id
+agim token bootstrap       # Print one-shot web-bootstrap token (`~/.agim/admin-bootstrap-token`)
+
 # (`im-hub-pro …` continues to work as a deprecated alias)
 ```
 
@@ -135,6 +161,11 @@ agim messengers            # List available messengers
 | `/outbox status\|list\|failed\|retry <id>` | Inspect & operate the persistent IM delivery queue (v1.1.2+) |
 | `/a2a stats\|recent\|tree <id>` | Agent-to-Agent observability — chain, latency, callee histogram (v1.1.2+) |
 | `/router status\|explain` | Inspect routing decisions |
+| `/router compare agent:<a> agent:<b> "<prompt>"` | A/B evaluator — run the same prompt against two backends, side-by-side card with latency / tokens / outputs (v1.3.0+) |
+| `/goal …` | Sustained per-thread objective — `set / body / progress / pause / resume / complete / cancel / list / show` (v1.3.0+) |
+| `/heartbeat …` | Periodic DECIDE→EXECUTE→EVALUATE tick — `bind / unbind / show / pause / resume / status` (v1.3.0+) |
+| `/skill list\|show <name>\|refresh` | Inspect agim Skills loaded from `~/.agim/skills/` (v1.3.0+) |
+| `/cc native …` / `/na …` / `/llm …` | Route to the in-process native LLM agent (v1.3.0+) |
 | `y` / `n` / `批准` / `拒绝` | Approve / deny Claude tool call (or reminder confirmation card) |
 | `1` / `2` | After a service restart, reply `1` to redo the interrupted message or `2` to cancel (10 min window) |
 
@@ -360,6 +391,7 @@ Environment="IMHUB_SMTP_PASS=xxxxxxxxxxxxxxxx"
 Single process, zero external dependencies — SQLite + session files are the entire persistence layer.
 
 For the full deep-dive see [`docs/architecture/current.md`](docs/architecture/current.md).
+For the native LLM stack and the peer-`AgentAdapter` design that lets CLI agents and the in-process loop share one router / approval bus / A2A / skills pipeline, see [`docs/architecture-bridge-and-native.md`](docs/architecture-bridge-and-native.md) and the PR-by-PR landing order in [`docs/nanobot-comparison-rollout.md`](docs/nanobot-comparison-rollout.md).
 
 ## Requirements
 
@@ -417,6 +449,8 @@ See [`docs/deployment.md`](docs/deployment.md) for systemd, Docker, nginx, monit
 | **v1.2.0** | **Long-term memory** (SQLite + FTS5 + persona + 4 MCP tools + auto-distillation + Web admin tab); **vector retrieval** (local BGE / OpenAI-compatible, RRF fusion); **Cost & Health dashboard**; **agent-initiated push** (`mcp__imhub__push_message`); recovery card with native inline buttons. |
 | v1.2.1 | Vector backend auto-warm on restart; Memory + Cost tab i18n. |
 | v1.2.2 | `agim admin invite` for onboarding additional admins via IM; vector eager-warm at boot; async consolidate (jobId + polling). |
+| v1.2.27–v1.2.32 | R14/R15 lifecycle hardening — 25s graceful-shutdown hard cap, port-first teardown, reap stray agim processes, argv-strict /proc scan in `findAgimProcesses`, PID file on every successful start, CLI wizard "Web bind+port" section, ACP port gate (`IMHUB_ENABLE_REMOTE_AGENT`). |
+| **v1.3.0** | **nanobot-vs-agim comparison rollout** — `native` in-process AgentAdapter (LLM client + agent loop + MCP attach + policy approval gate, OpenAI/Anthropic-compatible); `llmBackends` + `llmRoles` + `~/.agim/llm-secrets.json` config; `/goal` per-thread sustained objective (P0 #2); `/heartbeat` 3-phase scheduler (P0 #1); `mcp__imhub__ask_user` structured choice (P0 #5); agim Skills 3-tier engine (Stage 3 + P0 #3); inline row-age annotations on memory facts (P0 #4); notification evaluator for `mcp__imhub__push_message` (P0 #9); `/router compare` A/B evaluator (Stage 5); `tryIntrospectChain` fallback helper; Web admin pages `/settings/{llm,mcp,native-agent,agim-skills}` + `/tasks/{heartbeat,goals,asks}`; CLI wizards `agim setup {llm,mcp}` + diagnostics `agim diag {llm,mcp}` + `agim setup llm native-enable` quick path; CLI + Web full i18n (en/zh) for every new surface; `acp-token` separation from `web-token`; `agim token bootstrap` printer; `gen-endpoints --check` drift guard. See [`docs/nanobot-comparison-rollout.md`](docs/nanobot-comparison-rollout.md). |
 
 ### Next
 

package/README.zh-CN.md

@@ -13,7 +13,15 @@
 
 ## 亮点
 
-- **3 种 IM + 邮件，3+ 种 Agent** — 微信（图片 / 文件 / 语音）、飞书、钉钉（图片 / 语音，自带服务端 ASR）、Email（SMTP）；Claude Code、Codex、OpenCode，以及任意 ACP 端点
+- **3 种 IM + 邮件，3+ 种 Agent** — 微信（图片 / 文件 / 语音）、飞书、钉钉（图片 / 语音，自带服务端 ASR）、Email（SMTP）；Claude Code、Codex、OpenCode、Antigravity，以及任意 ACP 端点
+- **原生 LLM 栈（v1.3.0+，v1.2.48 强化）** — 内置 in-process agent loop（`native` adapter），直连任何 OpenAI / Anthropic 兼容 API（DeepSeek / Kimi / GLM / Qwen / Ollama / vLLM / OpenAI / Anthropic）。v1.2.48 让 native 真正进入生产可用：**角色定义** 通过 `~/.agim-workspaces/native/AGENTS.md`（热加载，无需重启）；**多模态**（图片附件 → openai-compat `image_url`）；**自动压缩**长对话（>60k 字符 → 由 cheap role 摘要）；**Provider 失败链路降级**（5xx/超时自动切下一个 role）；**轮级心跳**每 3 分钟一次"⏳ 还在处理"避免多步研究看似卡死；**agent-loop 超时拉到 28 分钟**（旧 5 分钟会切断 multi-hop `call_agent`）。可用作 `memory-distill` / `memory-consolidate` / `intent-llm` 的便宜路由，可作为本地优先的聊天 agent（`/na`），也可作 A2A 的 peer。详见 [`docs/llm-backends.md`](docs/llm-backends.md) 与 [`docs/architecture-bridge-and-native.md`](docs/architecture-bridge-and-native.md)
+- **Skills 多目录扫描（v1.2.48+）** — `IMHUB_SKILLS_MODE=auto`（默认）把 `~/.agim/skills/`（写入目标）与 `~/.claude/skills/`、`~/.config/opencode/skills/`、`~/.codex/skills/` 合并；新装即可看到所有已有 CLI agent 的技能。同名时 workspace 覆盖。Settings → 技能 改成两个子 tab（已安装编辑器 + skillhub 热门）；顶部导航条的 `/skills` 已合并进 Settings
+- **跨通道出口告警（v1.2.48+）** — 某条会话 outbox 失败 ≥3 次时（如 wechat session 过期），agim 自动失效 stale 的 context-token，并向你预设的备用通道（`IMHUB_OPERATOR_ALERT_PLATFORM` / `IMHUB_OPERATOR_ALERT_THREAD_ID`）推一条"⚠️ outbox 告警"，让你比用户更早发现
+- **会话级长目标（v1.3.0+ · `/goal`）** — 每个 IM 会话最多一个 active 目标，自动注入每条 prompt，确保 Agent 跨多轮不偏题。生命周期：`set / body / progress / pause / resume / complete / cancel / list`。跨会话管理在 `/tasks/goals`。详见 [`docs/goals.md`](docs/goals.md)
+- **Heartbeat 主动节拍（v1.3.0+ · `/heartbeat`）** — 把一个会话绑到周期性「现在有事要做吗？」节拍。每次 tick 走 DECIDE（便宜 LLM）→ EXECUTE（Agent 实际办）→ EVALUATE（通知门评估是否打扰用户）。跨会话管理在 `/tasks/heartbeat`
+- **结构化 Agent→User 选择题（v1.3.0+ · `mcp__imhub__ask_user`）** — Agent 给用户出 N 选 1 选择题，用户按键或文字回复都能可靠回到 Agent，避免「自由聊天里 yes/no 二义性」。等待队列在 `/tasks/asks`。详见 [`docs/ask-user.md`](docs/ask-user.md)
+- **agim Skills 技能引擎（v1.3.0+ · `/skill`）** — 放一个 `SKILL.md` 到 `~/.agim/skills/<name>/`，所有 CLI agent + native 都看得到。3 层渐进披露（name+desc 每轮自动注入；body 按需 `mcp__imhub__read_skill`；scripts/refs 走普通 file-read）。Web 编辑器在 `/settings/agim-skills`。详见 [`docs/skills.md`](docs/skills.md)
+- **`/router compare` A/B 评测（v1.3.0+）** — 同一条 prompt 同时打两个后端（CLI 与 native 可任意组合），一个卡片返回延迟 / token / 输出，方便在改 `defaultAgent` 之前先选个赢家
 - **合规友好默认（v1.2.23+）** — 海外 IM 适配器（Discord、Telegram）以及远程 ACP Agent 入口默认**全部隐藏**（CLI 向导 / Web 后台 / 运行时加载三层都 gate 住），主页面、安装向导、Web 后台都不会暴露。国内部署保持关闭即可。需要跨境通道时再通过 `IMHUB_ENABLE_GLOBAL_IM=1` / `IMHUB_ENABLE_REMOTE_AGENT=1` 显式启用。完整安全控制清单见 [`docs/security-hardening.md`](docs/security-hardening.md)（~30 个 env、敏感路径黑名单、出站脱敏、审计事件表、token 角色、R14 systemctl-aware 生命周期）
 - **Agent 之间互调（A2A）** — Agent 可以在自己的工具循环里直接调另一个 Agent，比如直接说「用 codex 帮我跑 git status」/「让 opencode 跑一下测试」。护栏在 agim 里强制：递归深度限制（默认 3）、禁止自调、工作区白名单、按人预算共享；可观测：`/a2a stats | recent | tree <id>`；`IMHUB_A2A_ENABLED=0` 关闭
 - **A2A 共享文件传递（v1.1.3+）** — caller 可通过 `inputs[]`（`fromAbsolutePath` / `fromCallerOutput` / 内联 `content`）把文件放进 callee 工作区；callee 把产物写到 `~/.agim/artifacts/<jobId>/_agim-output/`，返回的 `outputs[]` 让 caller 按需读。同文件系统优先 hard-link，否则复制。大小上限 env 可调
@@ -82,15 +90,18 @@ npm 包名是 `agim-cli`（短名 `agim` 被 npm 反 typo-squat 保留了）；
 | 分类 | 内容 |
 |------|------|
 | **IM 通道** | 微信（iLink — 图片 / 文件 / 语音 / 视频）、飞书（WebSocket）、钉钉（Stream 模式 — 图片 / 语音，自带服务端 ASR）、Email（SMTP，仅推送） |
-| **Agent** | Claude Code、Codex、OpenCode（统一 `AgentBase`）；任何 HTTP Agent 通过 ACP 接入 |
+| **Agent** | Claude Code、Codex、OpenCode、Antigravity（统一 `AgentBase`）；**`native`** in-process LLM loop；任何 HTTP Agent 通过 ACP 接入——所有后端走同一份 `AgentAdapter` 接口 |
+| **原生 LLM** | OpenAI / Anthropic 兼容 provider，外挂 MCP 服务器，多轮 agent loop + 策略式审批门，角色绑定（`cheap` / `evaluator` / `native-chat`） |
 | **提醒** | `/remind` slash、LLM 意图识别、LLM 润色投递、Agent MCP 工具（claude-code + opencode）、Web `/reminders` 页面 |
-| **Web 控制台** | 流式对话、三态主题（浅 / 深 / 跟随系统）、中英双语、SSE 实时仪表盘 |
-| **工具人审** | IM 端 + 页面内审批卡；Claude MCP sidecar |
+| **长目标 + 心跳 + 提问** | `/goal` 会话级目标、`/heartbeat` 周期 DECIDE/EXECUTE/EVALUATE 节拍、`mcp__imhub__ask_user` 结构化选择题，跨会话视图在 `/tasks/{goals,heartbeat,asks}` |
+| **技能（Skills）** | `~/.agim/skills/<name>/SKILL.md` 被所有 agent（CLI + native）发现；3 层渐进披露；Web 编辑器在 `/settings/agim-skills` |
+| **Web 控制台** | 流式对话、三态主题（浅 / 深 / 跟随系统）、中英双语（v1.3.0 每个新页面都跟齐）、SSE 实时仪表盘、Settings 子页 LLM / MCP / Native Agent / Skills |
+| **工具人审** | IM 端 + 页面内审批卡；Claude MCP sidecar；native 走 allow-list 策略门 |
 | **任务** | SQLite 持久化 Job Board + cron 调度；批量操作；后台任务读取 |
-| **路由** | 意图分类、断路器、限流、Sticky 会话、LLM 兜底 |
+| **路由** | 意图分类、断路器、限流、Sticky 会话、LLM 兜底；`/router compare` A/B 评测 |
 | **工作区** | 多租户；Agent 白名单 + 限流；命令级 ACL |
 | **可观测** | pino 结构化日志、traceId、Prometheus `/api/metrics`、SQLite 审计 |
-| **ACP** | 客户端（连接远端 Agent）+ 服务端（阿吉姆自身作为 ACP Agent） |
+| **ACP** | 客户端（连接远端 Agent）+ 服务端（阿吉姆自身作为 ACP Agent），bearer 走 `~/.agim/acp-token`（兼容回退 `~/.agim/web-token`） |
 
 ## CLI 命令
 
@@ -109,6 +120,21 @@ agim config claude         # 配置 Claude Code
 agim config agent          # 接入自定义 ACP Agent
 agim agents                # 列出可用 Agent
 agim messengers            # 列出可用 IM
+
+# v1.3.0+ — 原生 LLM 栈 + 诊断
+agim setup llm             # 向导：增/查/删 LLM backend，绑定角色，写 secret
+agim setup llm native-enable  # 一键启用 `native` agent（前提是已配置 backend）
+agim setup mcp             # 向导：增/查/删 外挂 MCP 服务器（stdio / HTTP / SSE）
+agim setup mcp health      # 探活全部 MCP 服务器
+agim diag llm              # 各 backend ping + 角色绑定表
+agim diag mcp              # 连接并列工具
+
+# v1.1.10+ — Web/ACP 令牌
+agim token create [name]   # 新发 user/admin token（接 /api/*、/acp/*）
+agim token list            # 列出 token（只回显末 4 位）
+agim token revoke <id>     # 按 id 撤销
+agim token bootstrap       # 输出一次性 web-bootstrap token（`~/.agim/admin-bootstrap-token`）
+
 # （`agim …` 作为别名继续可用）
 ```
 
@@ -127,6 +153,11 @@ agim messengers            # 列出可用 IM
 | `/outbox status\|list\|failed\|retry <id>` | 查看与重发 IM 投递队列（v1.1.2+） |
 | `/a2a stats\|recent\|tree <id>` | 查看 Agent 之间互调的统计 / 最近 N 次 / 调用链树（v1.1.2+） |
 | `/router status\|explain` | 查看路由策略 |
+| `/router compare agent:<a> agent:<b> "<内容>"` | A/B 评测——同一 prompt 同时打两个后端，并排展示延迟 / token / 输出（v1.3.0+） |
+| `/goal …` | 会话级长目标——`set / body / progress / pause / resume / complete / cancel / list / show`（v1.3.0+） |
+| `/heartbeat …` | 周期 DECIDE→EXECUTE→EVALUATE 节拍——`bind / unbind / show / pause / resume / status`（v1.3.0+） |
+| `/skill list\|show <name>\|refresh` | 查看从 `~/.agim/skills/` 加载的 agim Skills（v1.3.0+） |
+| `/cc native …` / `/na …` / `/llm …` | 切换到 in-process 原生 LLM agent（v1.3.0+） |
 | `y` / `n` / `批准` / `拒绝` | 同意 / 拒绝（工具调用 或 提醒确认卡片） |
 | `1` / `2` | 服务重启后看到「上次消息中断」提示时，回 `1` 重发 / `2` 取消（10 分钟内有效） |
 
@@ -352,6 +383,7 @@ Environment="IMHUB_SMTP_PASS=xxxxxxxxxxxxxxxx"
 单进程，零外部依赖——SQLite + 会话文件就是全部持久化层。
 
 深入架构详见 [`docs/architecture/current.md`](docs/architecture/current.md)。
+原生 LLM 栈与「CLI agent / in-process loop 都是 `AgentAdapter` 的 peer」的设计立场，详见 [`docs/architecture-bridge-and-native.md`](docs/architecture-bridge-and-native.md)；nanobot vs agim 比较计划的 PR 落地顺序见 [`docs/nanobot-comparison-rollout.md`](docs/nanobot-comparison-rollout.md)。
 
 ## 环境要求
 
@@ -409,6 +441,8 @@ systemd、Docker、nginx、监控、管理员管理、备份、升级详见 [`do
 | **v1.2.0** | **长期记忆**（SQLite + FTS5 + 记忆画像 + 4 个 MCP 工具 + 自动蒸馏 + Web 管理 Tab）；**向量召回**（本地 BGE / OpenAI 兼容、RRF 融合）；**成本与健康仪表盘**；**Agent 主动推送**（`mcp__imhub__push_message`）；带原生内联按钮的崩溃恢复卡片 |
 | v1.2.1 | 向量后端重启后自动加载；Memory + Cost tab 中文 i18n |
 | v1.2.2 | `agim admin invite` 通过 IM 邀请新管理员；向量后端 boot 时主动 warm；consolidate 改异步（jobId + 轮询） |
+| v1.2.27–v1.2.32 | R14/R15 生命周期加固——25s 优雅退出硬截止、端口先释放、reap stray agim 进程、`findAgimProcesses` argv-strict /proc 扫描、每次成功 start 都写 PID 文件、CLI 向导新增 "Web bind+port" 段、ACP 端口 gate（`IMHUB_ENABLE_REMOTE_AGENT`） |
+| **v1.3.0** | **nanobot vs agim 比较计划全量落地** —— `native` in-process AgentAdapter（LLM client + agent loop + 外挂 MCP + 策略式审批门，OpenAI/Anthropic 兼容）；`llmBackends` + `llmRoles` + `~/.agim/llm-secrets.json` 配置；`/goal` 会话级长目标（P0 #2）；`/heartbeat` 3 阶段调度（P0 #1）；`mcp__imhub__ask_user` 结构化选择题（P0 #5）；agim Skills 3 层引擎（Stage 3 + P0 #3）；记忆事实行内年龄注释（P0 #4）；`mcp__imhub__push_message` 的通知评估门（P0 #9）；`/router compare` A/B 评测（Stage 5）；`tryIntrospectChain` fallback 助手；Web 后台 `/settings/{llm,mcp,native-agent,agim-skills}` + `/tasks/{heartbeat,goals,asks}` 全新页面；CLI 向导 `agim setup {llm,mcp}` + 诊断 `agim diag {llm,mcp}` + `agim setup llm native-enable` 快捷通道；CLI + Web 所有新增面板的中英 i18n；`acp-token` 与 `web-token` 分离；`agim token bootstrap` 输出器；`gen-endpoints --check` drift guard。详见 [`docs/nanobot-comparison-rollout.md`](docs/nanobot-comparison-rollout.md) |
 
 ### 接下来
 

package/dist/cli-ui/config-wizard.js

@@ -472,18 +472,31 @@ async function configureDiscord(lang, cfg) {
     };
 }
 const AGENT_SPECS = [
+    { id: 'native', cmd: '__native__' },
     { id: 'claude-code', cmd: 'claude' },
     { id: 'opencode', cmd: 'opencode', authHintKey: 'agents.auth_hint.opencode' },
     { id: 'codex', cmd: 'codex', authHintKey: 'agents.auth_hint.codex' },
 ];
+async function nativeLlmConfigured() {
+    try {
+        const cfg = await loadConfig();
+        const raw = cfg;
+        return Array.isArray(raw.llmBackends) && raw.llmBackends.length > 0;
+    }
+    catch {
+        return false;
+    }
+}
 /** Detect install state once per sectionAgents() call, but always
  *  re-detect after the user picks "Re-check". */
 async function detectAgentInstalls() {
     const out = new Map();
     await Promise.all(AGENT_SPECS.map(async (spec) => {
-        {
-            out.set(spec.id, await cliAvailable(spec.cmd));
+        if (spec.id === 'native') {
+            out.set(spec.id, await nativeLlmConfigured());
+            return;
         }
+        out.set(spec.id, await cliAvailable(spec.cmd));
     }));
     return out;
 }