feat(local): use a Claude plan in chat and Agentic mode

.agents/skills/translate/glossary.md

@@ -71,9 +71,10 @@ SurfSense, Studio, Markdown, llama.cpp, GGUF, Hugging Face, API, MCP, and every
 | session log | セッションログ | 세션 로그 | 会话日志 | सेशन लॉग | Sitzungsprotokoll | registro de la sesión | journal de session | log da sessão | журнал сеанса |
 | clipboard | クリップボード | 클립보드 | 剪贴板 | क्लिपबोर्ड | Zwischenablage | portapapeles | presse-papiers | área de transferência | буфер обмена |
 | embedding (the model that turns the library into vectors) | 埋め込み | 임베딩 | 嵌入 | एम्बेडिंग | Embedding | embedding | embedding | embedding | эмбеддинг |
-| sign in (to a ChatGPT account) | サインイン | 로그인 | 登录 | साइन इन | anmelden | iniciar sesión | se connecter | entrar | войти |
-| plan (a ChatGPT plan, Plus or Pro) | プラン | 플랜 | 套餐 | प्लान | Plan | plan | forfait | plano | план |
+| sign in (to a ChatGPT account, or Claude Code to a Claude plan) | サインイン | 로그인 | 登录 | साइन इन | anmelden | iniciar sesión | se connecter | entrar | войти |
+| plan (a ChatGPT plan, Plus or Pro; a Claude plan, Pro, Max or Team) | プラン | 플랜 | 套餐 | प्लान | Plan | plan | forfait | plano | план |
 | subscription (using a ChatGPT plan instead of a key) | サブスクリプション | 구독 | 订阅 | सब्सक्रिप्शन | Abo | suscripción | abonnement | assinatura | подписка |
+| terminal (where the user runs `claude`, the command that stays in English) | ターミナル | 터미널 | 终端 | टर्मिनल | Terminal | terminal (una) | terminal | terminal | терминал |
 
 ## What the English means
 

docs/adr/0057-a-claude-plan-runs-through-the-users-own-claude-code.md

@@ -1,6 +1,6 @@
 # ADR 0057: A Claude plan is reached through the user's own Claude Code, and SurfSense holds no Claude credential
 
-- **Status:** Accepted
+- **Status:** Accepted; its network decision superseded by [ADR 0061](0061-claude-code-reaches-anthropic-and-in-the-agent-the-turns-plugins.md)
 - **Date:** 2026-10-16
 - **Supersedes:** [ADR 0038](0038-chatgpt-plans-sign-in-through-openai-not-codex.md) in part: its consequence that a Claude subscription is not offered. [ADR 0050](0050-a-model-answers-on-its-own-route-with-its-connections-credential.md) in part: that SurfSense owns a plan's sign-in for both engines, which now holds for the ChatGPT plan only
 - **Source:** [Use your Claude plan proposal, decisions and Anthropic's terms L59–84](https://github.com/MODSetter/SurfSense/blob/e1332966dd2ec2cfcf636331da6f770d836fcff1/docs/proposals/claude-plan/README.md#L59-L84)
@@ -33,4 +33,4 @@ A Claude plan has no HTTP endpoint SurfSense may call, so opencode cannot reach
 
 ## Where the code stands
 
-Not built. The work is laid out in the [Use your Claude plan](../proposals/claude-plan/README.md) proposal.
+Built: chat, titles, Studio and Agentic mode ([Claude plan](../architecture/claude-plan.md)). It is untested on Windows and Linux on a real install. The contract tests against the real binary that the consequences name run only on a developer's machine, opted in; a Claude Code newer than the one they last passed on is named in the log, and the version SurfSense calls tested is raised only once they pass again ([Tests](../architecture/claude-plan.md#tests), [Known gaps](../architecture/claude-plan.md#known-gaps)).

docs/adr/0060-one-consent-question-per-user-action.md

@@ -0,0 +1,25 @@
+# ADR 0060: One consent question asks for every host a user action needs, including hosts its background work will reach
+
+- **Status:** Accepted
+- **Date:** 2026-10-16
+
+## Context
+
+Egress consent is per host ([ADR 0027](0027-egress-consent-per-host.md)) and asked just in time, on a user action, never on a background call ([ADR 0017](0017-egress-off-by-default.md)). Some actions need several hosts: a ChatGPT sign-in reaches OpenAI's sign-in host and its API; a plugin can sign in on a host of its own; a Claude plan's Claude Code calls `api.anthropic.com` and refreshes its sign-in on `platform.claude.com`. The prompt asked about one host at a time, so a user could allow one and cancel the next, leaving a connection that cannot work.
+
+A host one action needs is not always reached during it. Claude Code refreshes its sign-in in the background, whenever its token is near expiry, at the start of any run; the first run is the model list right after connecting. No dialog can be raised then, so a host left to be asked later is a refusal the user never sees: their Claude plan stops answering once its token expires.
+
+## Decision
+
+- One question per user action, naming every host that action may reach, including a host that work started by it reaches in the background, since that host can never be asked about later.
+- A host reached only by a later, separate user action is asked about then, in its own question.
+- One Allow switches every host the question named, in one request committed whole (`PUT /egress`): all are allowed or none is.
+- Consent stays per host: each is its own row in Settings › Network, switched off alone, and checked on every call.
+- The question says what each host receives when they differ: a Claude plan's API host gets what is sent to the model, its sign-in host only the sign-in's refresh.
+- Only hosts SurfSense names are ever asked about. A host a program SurfSense runs tries on its own, such as a new one in a future Claude Code, is refused until a release adds it, reviewed and described, to that program's list.
+
+## Consequences
+
+- Connecting a Claude plan, signing in with ChatGPT, or connecting a plugin with its own sign-in host is one question.
+- A user cannot allow part of what one action needs from the question itself; they can turn any host off afterwards in Settings › Network.
+- A feature whose background work reaches a host its own action did not name fails closed: the host is refused, and the refusal is logged.

docs/adr/0061-claude-code-reaches-anthropic-and-in-the-agent-the-turns-plugins.md

@@ -0,0 +1,26 @@
+# ADR 0061: Claude Code reaches Anthropic's two hosts, and in the agent also the hosts of the turn's ready plugins
+
+- **Status:** Accepted
+- **Date:** 2026-10-16
+
+## Context
+
+[ADR 0057](0057-a-claude-plan-runs-through-the-users-own-claude-code.md) has Claude Code's traffic tunnel "only `api.anthropic.com`". Claude Code also refreshes its sign-in on `platform.claude.com`, so chat runs already allow both, each while the user allows it ([ADR 0060](0060-one-consent-question-per-user-action.md)).
+
+The agent gives Claude Code the thread's plugins. opencode connects to each plugin server itself, with its URL and sign-in header, through the plugin proxy, which passes only a ready plugin's hosts ([agent](../architecture/agent.md#network)). Measured on Claude Code 2.1.301: its MCP traffic to a remote plugin goes through `HTTPS_PROXY`, so a proxy limited to Anthropic refuses the plugin.
+
+The alternative was for SurfSense's tool endpoint to list and forward plugin tools, so Claude Code would reach only Anthropic. SurfSense has no code that forwards an MCP call; it would be a second MCP client beside opencode's, with its own sessions, timeouts and cancellation, to keep plugin tokens out of a process that opencode already gives them to.
+
+## Decision
+
+- A chat run's Claude Code reaches `api.anthropic.com` and `platform.claude.com`, each while its destination is on, and nothing else. A chat run is given no MCP server.
+- An agent run's Claude Code reaches the same two hosts and the hosts of the ready plugins, by the check the plugin proxy applies to opencode. It connects to each plugin server itself, as opencode does.
+- The two run kinds use separate proxies, so a chat run can never reach a plugin host.
+- The MCP servers, with their sign-in headers, are passed to Claude Code in a file only the user can read, never on its command line.
+
+## Consequences
+
+- Claude Code reaches no host the user has not allowed for the same purpose in opencode.
+- Claude Code holds the sign-in headers of the plugins it is given for the life of its process, as opencode does.
+- A change to how the plugin proxy decides readiness applies to both engines.
+- This supersedes ADR 0057's "tunnels only `api.anthropic.com`".

docs/adr/README.md

@@ -84,6 +84,8 @@ The Date is the day the decision was first written down. Source links are permal
 | 0054 | [SurfSense's own plugins, free and paid, run on one plugin host, with the registry served beside it](0054-surfsenses-plugins-run-on-one-plugin-host.md) | Accepted |
 | 0055 | [Verified is a maintainer's judgement, and CI contacts each server a pull request adds](0055-verified-is-a-maintainers-judgement-and-ci-contacts-each-server.md) | Accepted |
 | 0056 | [Speech recognition runs on a second audio.cpp server that only the API calls](0056-speech-recognition-runs-on-a-second-audiocpp-server.md) | Accepted |
-| 0057 | [A Claude plan is reached through the user's own Claude Code, and SurfSense holds no Claude credential](0057-a-claude-plan-runs-through-the-users-own-claude-code.md) | Accepted |
+| 0057 | [A Claude plan is reached through the user's own Claude Code, and SurfSense holds no Claude credential](0057-a-claude-plan-runs-through-the-users-own-claude-code.md) | Accepted, its network decision superseded by 0061 |
 | 0058 | [The bundled embedder runs at FP32 on x86-64, upcast in memory, and its vector space is still the pinned FP16 file](0058-the-bundled-embedder-runs-at-fp32-on-x86-64.md) | Accepted |
 | 0059 | [A born-digital PDF is read from its own text layer, and Docling reads only a PDF with a page that needs it](0059-a-born-digital-pdf-is-read-from-its-text-layer.md) | Accepted |
+| 0060 | [One consent question asks for every host a user action needs, including hosts its background work will reach](0060-one-consent-question-per-user-action.md) | Accepted |
+| 0061 | [Claude Code reaches Anthropic's two hosts, and in the agent also the hosts of the turn's ready plugins](0061-claude-code-reaches-anthropic-and-in-the-agent-the-turns-plugins.md) | Accepted |