@markusylisiurunen/tau 0.3.18 → 0.3.20

package/README.md

@@ -82,10 +82,10 @@ filters: `--since`, `--persona`, `--provider`, `--model`.
 tau can run without the TUI via NDJSON RPC over stdin/stdout:
 
 ```sh
-tau rpc --persona gpt-5.5-coder --risk read-only
+tau rpc --persona gpt-5.5-coder
 ```
 
-RPC mode reuses the same startup config, persona loading, and risk handling as interactive mode. stdin/stdout are reserved for protocol traffic in this mode (piped stdin is **not** treated as an initial user message). `--caffeinated` is a macOS-only TUI flag and is rejected outside TUI mode.
+RPC mode reuses the same startup config and persona loading as interactive mode. stdin/stdout are reserved for protocol traffic in this mode (piped stdin is **not** treated as an initial user message). `--caffeinated` is a macOS-only TUI flag and is rejected outside TUI mode.
 
 for protocol details and examples, see [docs/rpc.md](docs/rpc.md).
 
@@ -94,7 +94,7 @@ for protocol details and examples, see [docs/rpc.md](docs/rpc.md).
 tau can host sessions over WebSocket:
 
 ```sh
-tau serve --host 0.0.0.0 --port 8787 --auth-token "$TAU_WS_AUTH_TOKEN" --risk read-only
+tau serve --host 0.0.0.0 --port 8787 --auth-token "$TAU_WS_AUTH_TOKEN"
 ```
 
 WebSocket auth tokens authorize full session access. Prefer `wss://` behind a trusted TLS proxy on untrusted networks, avoid putting tokens in URLs or shell history, and treat any proxy/access logs that capture headers, query strings, or WebSocket handshake details as sensitive.
@@ -127,16 +127,16 @@ tau attach --new --execution-kind fly-sprite --fly-api default --fly-sprite spri
 tau can also run the terminal UI against any command that speaks the same session protocol on stdin/stdout:
 
 ```sh
-tau attach -- ssh vps 'cd /path/to/repo && tau rpc --risk read-only'
+tau attach -- ssh vps 'cd /path/to/repo && tau rpc'
 ```
 
 For stdio attach, use `--session <id>` before `--`:
 
 ```sh
-tau attach --session 0195d6e4-4cf9-7f44-a2d8-f8f7f49ee9d3 -- ssh vps 'cd /path/to/repo && tau rpc --risk read-only'
+tau attach --session 0195d6e4-4cf9-7f44-a2d8-f8f7f49ee9d3 -- ssh vps 'cd /path/to/repo && tau rpc'
 ```
 
-Session attach renders the authoritative session snapshot, streams recoverable `session.delta` updates and independently revisioned, non-persisted `session.pendingUserMessages` replacements, submits normal user input through `session.submit`, `session.queue`, and `session.steer`, supports steering/interruption, runs `!`/`!!` shell commands in the session execution environment, records `/listen` from the local microphone, speaks `/speak` locally, reloads session content with `/reload`, changes the session risk level with `/risk:<level>` or `Ctrl+R`, switches session personas with `/persona:<id>` or `Ctrl+P`, inserts session prompt templates with `/prompt:<id>`, compacts or prunes the session with `/compact:*` and `/prune:*`, creates a new session with `/new`, and exits with `/exit` or `Ctrl+C` twice.
+Session attach renders the authoritative session snapshot, streams recoverable `session.delta` updates and independently revisioned, non-persisted `session.pendingUserMessages` replacements, submits normal user input through `session.submit`, `session.queue`, and `session.steer`, supports steering/interruption, runs `!`/`!!` shell commands in the session execution environment, records `/listen` from the local microphone, speaks `/speak` locally, reloads session content with `/reload`, switches session personas with `/persona:<id>` or `Ctrl+P`, inserts session prompt templates with `/prompt:<id>`, compacts or prunes the session with `/compact:*` and `/prune:*`, creates a new session with `/new`, and exits with `/exit` or `Ctrl+C` twice.
 
 ## Telegram runner
 
@@ -150,11 +150,12 @@ The Telegram command surface is intentionally small:
 
 - `/new` creates a new session for the chat, replacing the previous active session if one exists.
 - `/status` returns a short natural-language status paragraph for the active session, including model, reasoning effort, context usage, and cumulative cost when session details are available.
+- `/compact` summarizes older conversation context to reduce context usage.
 - `/interrupt` interrupts the active run.
 
 The runner keeps tool and lifecycle progress quiet: Telegram receives command acknowledgements/errors and assistant messages, including multiple assistant progress updates from a single active run. While work is active, it shows Telegram's typing indicator in DMs and groups. Assistant messages are sent as Telegram rich markdown.
 
-Telegram DM input supports plain text, voice/audio transcription, and attachment queueing (`image/*`, PDF, `.txt`, `.md`, `.json`, `.csv`, `.yaml`, `.yml`). allowed groups are opt-in via `allowedChatIds`; non-triggering group text/captions, attachments, audio transcripts, and processing errors are buffered as sender-attributed context and the most recent 50 messages since the previous bot-triggering turn are included when a bot mention triggers a turn. group commands accept explicit bot mentions on or around the command, such as `/status@botusername`, `/status @botusername`, or `@botusername /status`.
+Telegram DM input supports plain text, voice/audio transcription with the transcript echoed back for verification, and attachment queueing (`image/*`, PDF, `.txt`, `.md`, `.json`, `.csv`, `.yaml`, `.yml`). allowed groups are opt-in via `allowedChatIds`; non-triggering group text/captions, attachments, audio transcripts, and processing errors are buffered as sender-attributed context and the most recent 50 messages since the previous bot-triggering turn are included when a bot mention triggers a turn. group commands accept explicit bot mentions on or around the command, such as `/status@botusername`, `/status @botusername`, or `@botusername /status`.
 
 Telegram config defines `bots`, `projects`, `workspaceRoot`, optional `systemMessage`, and optional `maxSessions`. repositories use persistent bare caches at `<workspaceRoot>-repo-cache/<projectId>.git`; active session records are persisted at `<workspaceRoot>-sessions.json`, their workspaces survive runner restarts, and unreferenced workspace entries are removed on startup. Tau reconnects preserved sessions to their corresponding snapshots. When creating or reconstructing a workspace, `bootstrapCommands` block initial readiness and `backgroundBootstrapCommands` start without blocking; preserved workspaces skip both on restart. for config details, see [docs/telegram.md](docs/telegram.md).
 
@@ -180,6 +181,7 @@ tau nook setup \
   --access-aud <access-application-audience>
 tau nook deploy ./dist --site demo
 tau nook deploy ./dist --site demo --public
+mkdir restored-demo && tau nook copy demo ./restored-demo
 tau nook template save starter ./app
 tau nook template copy starter ./next-app
 tau nook kv put demo settings '{"theme":"dark"}'
@@ -197,7 +199,7 @@ Add a single configured target to Tau config after deploying the Worker and crea
 }
 ```
 
-Template copies require a destination directory that already exists and is empty. The Worker validates Cloudflare Access JWTs against the Access JWKS with the configured issuer and audience. Tau sends service-token headers to Cloudflare Access for CLI/API calls, but the Worker does not treat those raw headers as authentication. When `nook` is configured, Tau automatically exposes a read-write model tool named `nook`. Detailed setup, deploy, template, Worker, browser SDK, and V0 scope notes live in [src/nook/README.md](src/nook/README.md).
+Template copies require a destination directory that already exists and is empty. The Worker validates Cloudflare Access JWTs against the Access JWKS with the configured issuer and audience. Tau sends service-token headers to Cloudflare Access for CLI/API calls, but the Worker does not treat those raw headers as authentication. When `nook` is configured, Tau automatically exposes a model tool named `nook`. Detailed setup, deploy, template, Worker, browser SDK, and V0 scope notes live in [src/nook/README.md](src/nook/README.md).
 
 ## SDK usage (Node)
 
@@ -256,13 +258,8 @@ this writes prompts and skills into `.tau/` under your current working directory
 
 ## security notice
 
-**the risk level system is a UX guardrail, not a hard security boundary.** it helps prevent accidental writes and guides model behavior, but it has significant limitations:
-
-- **model trust**: the bash tool relies on the model honestly declaring whether a command is a read or write. there's no runtime validation that the command actually matches the declared intent. a model could declare `safetyLevel="read"` while running `rm -rf /`.
-- **no command analysis**: the system doesn't inspect command content. it trusts the declared safety level without verifying what the command actually does.
 - **full system access**: the model can access any file on your system that your user account can read or write, not just the current working directory. if you need stronger isolation, run Tau inside a VM or container.
 - **no tty / non-interactive tools**: tool commands run with stdin ignored and no TTY. anything that prompts for input or opens an editor can hang or fail (for example `sudo`, `ssh` password prompts, `git` credential prompts). tau also forces git into non-interactive mode (no prompt/editor/pager, batch-mode ssh).
-- **user bypasses**: the `!` prefix executes shell commands directly and completely bypasses risk level checks. this is intentional for direct use, but means risk levels only constrain the model, not the user.
 
 note that there is no confirmation step before tool execution. the model runs commands immediately, and you can only observe the results after the fact.
 
@@ -303,7 +300,6 @@ available palette tokens (theme keys):
 - diff: `diffAdd`, `diffRemove`
 - toasts: `toastSuccess`, `toastWarn`, `toastError`
 - user: `userSurface`, `userMemorySurface`, `userMemoryText`
-- risk: `riskReadOnlyText`, `riskReadWriteText`
 
 example theme file (`.tau/themes/solarized.json`):
 
@@ -321,24 +317,9 @@ and in config (`.tau/config.json` or `~/.config/tau/config.json`):
 { "defaultTheme": "solarized" }
 ```
 
-## risk levels
-
-tau uses risk levels to control what the model can do. this lets you stay in control while working alongside AI.
-
-- **read-only** (default): model can run read-only tools (no file modifications)
-- **read-write**: model can create, edit, and delete files
-
-sub-agents inherit the session risk level unless overridden in persona config. a sub-agent configured with `riskLevel: read-write` can write even when the main session is `read-only`.
-
-start with a specific risk level (exact values: `read-only` or `read-write`):
-
-```sh
-tau --risk read-write
-```
-
-or change it during a session with `/risk:read-only` or `/risk:read-write`.
+## tool access
 
-the default is read-only because it lets the model investigate your code and answer questions without risk of unintended changes. bump it to read-write when you're ready to let the model make edits.
+enabled tools execute directly. persona and sub-agent tool lists determine which tools are available.
 
 ## power management (macOS)
 
@@ -357,10 +338,10 @@ tau comes with several built-in personas across different models:
 - **Claude Opus 4.6** (Anthropic): `opus-4.6-chat`, `opus-4.6-coder`
 - **Claude Opus 4.8** (Anthropic): `opus-4.8-chat`, `opus-4.8-coder`
 - **GPT-5.5** (OpenAI): `gpt-5.5-chat`, `gpt-5.5-coder`
-- **GPT-5.5 Fast (ChatGPT)** and **GPT-5.6 Fast (ChatGPT)** (OpenAI Codex, priority tier): `gpt-5.5-chatgpt-fast-chat`, `gpt-5.5-chatgpt-fast-coder`, `gpt-5.6-sol-chatgpt-fast-chat`, `gpt-5.6-sol-chatgpt-fast-coder`, `gpt-5.6-terra-chatgpt-fast-chat`, `gpt-5.6-terra-chatgpt-fast-coder`
+- **GPT-5.5 Fast (ChatGPT)** and **GPT-5.6 Fast (ChatGPT)** (OpenAI Codex, priority tier): `gpt-5.5-chatgpt-fast-chat`, `gpt-5.5-chatgpt-fast-coder`, `gpt-5.6-sol-chatgpt-fast-chat`, `gpt-5.6-sol-chatgpt-fast-coder`, `gpt-5.6-terra-chatgpt-fast-chat`, `gpt-5.6-terra-chatgpt-fast-coder`, `gpt-5.6-luna-chatgpt-fast-chat`, `gpt-5.6-luna-chatgpt-fast-coder`
 - **GPT-5.5 (ChatGPT)** (OpenAI Codex): `gpt-5.5-chatgpt-chat`, `gpt-5.5-chatgpt-coder`
-- **GPT-5.6 Sol** and **GPT-5.6 Terra** (OpenAI): `gpt-5.6-sol-chat`, `gpt-5.6-sol-coder`, `gpt-5.6-terra-chat`, `gpt-5.6-terra-coder`
-- **GPT-5.6 Sol (ChatGPT)** and **GPT-5.6 Terra (ChatGPT)** (OpenAI Codex): `gpt-5.6-sol-chatgpt-chat`, `gpt-5.6-sol-chatgpt-coder`, `gpt-5.6-terra-chatgpt-chat`, `gpt-5.6-terra-chatgpt-coder`
+- **GPT-5.6 Sol**, **GPT-5.6 Terra**, and **GPT-5.6 Luna** (OpenAI): `gpt-5.6-sol-chat`, `gpt-5.6-sol-coder`, `gpt-5.6-terra-chat`, `gpt-5.6-terra-coder`, `gpt-5.6-luna-chat`, `gpt-5.6-luna-coder`
+- **GPT-5.6 Sol (ChatGPT)**, **GPT-5.6 Terra (ChatGPT)**, and **GPT-5.6 Luna (ChatGPT)** (OpenAI Codex): `gpt-5.6-sol-chatgpt-chat`, `gpt-5.6-sol-chatgpt-coder`, `gpt-5.6-terra-chatgpt-chat`, `gpt-5.6-terra-chatgpt-coder`, `gpt-5.6-luna-chatgpt-chat`, `gpt-5.6-luna-chatgpt-coder`
 - **Gemini 3.1 Pro** and **Gemini 3 Flash** (Google): `gemini-3.1-pro-chat`, `gemini-3-flash-chat`
 
 chat variants are for general-purpose assistance; coder variants are optimized for software engineering. built-in personas include the `default` sub-agent for background tasks unless disabled.
@@ -377,9 +358,9 @@ tau --persona opus-4.8-coder
 
 some personas can run isolated sub-agents via the `spawn_agent`, `send_input_to_agent`, `wait_for_agents`, and `terminate_agent` tools. sub-agents report progress in the subagent panel, and `wait_for_agents` returns completed outputs as soon as at least one requested agent finishes.
 
-the built-in `default` sub-agent is available unless disabled. it inherits the main persona's model, settings, tool access (minus sub-agent management tools), risk level, and system prompt. the inherited main prompt is wrapped with default sub-agent-specific rules, and those wrapper rules take precedence on conflicts. custom sub-agents can override model, reasoning, tools, and risk level. a sub-agent configured with `riskLevel: read-write` can perform writes even when the main session is `read-only`.
+the built-in `default` sub-agent is available unless disabled. it inherits the main persona's model, settings, tool access (minus sub-agent management tools), and system prompt. the inherited main prompt is wrapped with default sub-agent-specific rules, and those wrapper rules take precedence on conflicts. custom sub-agents can override model, reasoning, and tools.
 
-`spawn_agent` supports an optional launch override string (`model: "<provider>/<model>:<effort>"`) and an optional `workingDirectory`. when `workingDirectory` is set, the sub-agent runs from that directory and its prompt context (cwd, AGENTS.md scope, and skills block) is rebuilt as if tau was started there. launch overrides are allowlisted per subagent. custom subagents can define `launchModels` in persona frontmatter, and the built-in `default` sub-agent uses `subagents.defaultLaunchModels` from config.
+`spawn_agent` supports an optional launch override string (`model: "<provider>/<model>:<effort>"`) and an optional `workingDirectory`. when `workingDirectory` is set, the sub-agent runs from that directory and its config, model catalog, repository metadata, AGENTS.md context, and skills are resolved through the session execution environment as if tau was started there. launch overrides are allowlisted per subagent. custom subagents can define `launchModels` in persona frontmatter, and the built-in `default` sub-agent uses `subagents.defaultLaunchModels` from config.
 
 sub-agent progress appears in a sticky panel. use `alt+down` to cycle active subagents and `ctrl+g` to terminate the selected one. tau caps active subagents at 8.
 
@@ -467,9 +448,7 @@ tau supports slash commands for common actions:
 | `/persona:<id>` | switch to a different persona |
 | `/prompt:<id>` | insert a saved prompt template |
 | `/theme:<id>` | switch to a loaded theme |
-| `/risk:read-only` | allow read-only tool calls |
-| `/risk:read-write` | allow all tools |
-| `!<cmd>` | run a shell command directly (bypasses risk checks) |
+| `!<cmd>` | run a shell command directly |
 | `!!<cmd>` | run a shell command without adding output to the model context |
 
 tau automatically compacts long sessions when provider-reported usage approaches the model context limit. automatic compaction summarizes older context, asks the compaction model to select original user messages to append verbatim inside the summary by history id, retains a recent tail verbatim, and inserts a hidden continuation note so the assistant continues without asking you to repeat context.
@@ -491,7 +470,6 @@ the prune commands drop bash tool results from the active context without summar
 | key          | action                                    |
 | ------------ | ----------------------------------------- |
 | `shift+tab`  | cycle reasoning effort                    |
-| `ctrl+r`     | cycle risk level                          |
 | `ctrl+p`     | cycle personality                         |
 | `ctrl+t`     | toggle thinking visibility                |
 | `ctrl+o`     | toggle compact tool display               |
@@ -524,7 +502,6 @@ model definitions can be extended and overridden through `~/.config/tau/models.j
     "mistral": "..."
   },
   "defaultPersona": "gpt-5.5-chat",
-  "defaultRisk": "read-write",
   "disableBuiltinPersonas": false,
   "disableBuiltinThemes": false,
   "defaultTheme": "solarized",
@@ -578,8 +555,6 @@ for built-in providers and features, the `apiKeys` field uses these keys: `anthr
 
 the `defaultPersona` field specifies which persona to use when starting the app. it accepts `<id>` or `<id>:<reasoning>`, and matching is exact/case-sensitive. the `--persona` flag overrides this setting.
 
-the `defaultRisk` field sets the initial risk level (`read-only` or `read-write`). the `--risk` flag overrides this setting. if not specified, defaults to `read-only`.
-
 the `defaultTheme` field sets the theme id to load at startup. it must be non-empty, and matching is exact/case-sensitive. if not specified, it defaults to `gold`.
 
 `speechToText.provider` selects the `/listen` and Telegram audio transcription provider. supported values are `mistral` (default, uses Voxtral) and `gemini` (uses Gemini 3.5 Flash with minimal thinking).
@@ -673,8 +648,8 @@ optional frontmatter fields:
 - `serviceTier`: `priority` or `flex` for providers that support service tiers (currently `openai` and `openai-codex`)
 - `allowedReasoningLevels`: list of reasoning levels shown in the ui
 - `skills`: list of enabled skill names (matched by `name` in skill frontmatter), or `"*"` to enable all discovered skills. if omitted, custom personas default to `"*"`. set `skills: []` to disable skills completely.
-- `tools`: optional list of persona-selected host tools. Nook is not selected here; when effective config contains `nook`, Tau exposes the `nook` tool automatically and gates every operation on read-write risk.
-- `subagents`: optional map of subagent definitions. the built-in `default` sub-agent is implicit unless `default: false` is provided. custom subagents must include `systemPrompt` and may include `description`, `provider`+`model`, `reasoning`, `serviceTier`, `tools`, `riskLevel`, and `launchModels` (when specifying a model, `provider` and `model` must be provided together). names must be lowercase with dashes (max 64 chars). `launchModels` entries must use `<provider>/<model>:<effort>` and are used to allowlist launch-time `spawn_agent` overrides. example:
+- `tools`: optional list of persona-selected host tools. Nook is not selected here; when effective config contains `nook`, Tau exposes the `nook` tool automatically.
+- `subagents`: optional map of subagent definitions. the built-in `default` sub-agent is implicit unless `default: false` is provided. custom subagents must include `systemPrompt` and may include `description`, `provider`+`model`, `reasoning`, `serviceTier`, `tools`, and `launchModels` (when specifying a model, `provider` and `model` must be provided together). names must be lowercase with dashes (max 64 chars). `launchModels` entries must use `<provider>/<model>:<effort>` and are used to allowlist launch-time `spawn_agent` overrides. example:
   ```yaml
   subagents:
     default: false
@@ -686,12 +661,11 @@ optional frontmatter fields:
       model: claude-haiku-4-5
       reasoning: medium
       tools: [web_search, web_fetch, bash]
-      riskLevel: read-only
       launchModels:
         - openai/gpt-5.5:high
         - anthropic/claude-haiku-4-5:medium
   ```
-- `tools`: list of tool names to enable for this persona. allowed: `bash`, `write`, `edit`, `view_image`, `spawn_agent`, `send_input_to_agent`, `wait_for_agents`, `terminate_agent`. if omitted, defaults to `bash`, `write`, `edit`, `view_image` (and subagent tools when subagents are enabled). risk levels still apply.
+- `tools`: list of tool names to enable for this persona. allowed: `bash`, `write`, `edit`, `view_image`, `spawn_agent`, `send_input_to_agent`, `wait_for_agents`, `terminate_agent`. if omitted, defaults to `bash`, `write`, `edit`, `view_image` (and subagent tools when subagents are enabled).
 
 the markdown body becomes the system prompt.
 
@@ -743,7 +717,7 @@ use `/reload` to pick up changes to personas, model overrides, prompts, skills,
 
 ## how it works
 
-tau connects your terminal to large language models, giving them tools to interact with your filesystem. when you ask the model to explore code or make changes, it decides which tools to use and executes them subject to the active risk level.
+tau connects your terminal to large language models, giving them tools to interact with your filesystem. when you ask the model to explore code or make changes, it decides which tools to use and executes them directly.
 
 the model sees your messages, any file contents you've shared, and the results of tool calls. it doesn't have ambient access to your filesystem; it only sees what you show it or what it explicitly requests through tools.
 

package/dist/core/cli.js

@@ -1,5 +1,5 @@
 import { parsePersonaReference } from "./persona_reference.js";
-import { REASONING_LEVELS, RiskLevelSchema, } from "./types.js";
+import { REASONING_LEVELS } from "./types.js";
 export class CliError extends Error {
     constructor(message) {
         super(message);
@@ -23,7 +23,7 @@ export function parsePersonaString(raw, personas) {
         reasoning: personaId ? parsed.reasoning : undefined,
     };
 }
-function parsePersonaOption(raw, personas) {
+function parsePersonaOption(raw) {
     const parsedReference = parsePersonaReference(raw);
     if (parsedReference.error === "missing-reasoning") {
         throw new CliError("missing reasoning level after ':' in --persona");
@@ -35,28 +35,11 @@ function parsePersonaOption(raw, personas) {
     if (!parsedReference.personaId) {
         throw new CliError("missing persona id in --persona");
     }
-    const personaId = resolvePersonaId(parsedReference.personaId, personas);
-    if (!personaId) {
-        const available = personas.map((p) => p.id).join(", ");
-        throw new CliError(`unknown persona '${parsedReference.personaId}'. available personas: ${available}`);
-    }
     return {
-        personaId,
+        personaId: parsedReference.personaId,
         reasoningOverride: parsedReference.reasoning,
     };
 }
-function parseRiskOption(raw) {
-    const normalized = raw.trim();
-    if (!normalized) {
-        throw new CliError("missing value for --risk");
-    }
-    const parsed = RiskLevelSchema.safeParse(normalized);
-    if (!parsed.success) {
-        const allowed = RiskLevelSchema.options.join(", ");
-        throw new CliError(`invalid risk level '${raw}'. allowed levels: ${allowed}`);
-    }
-    return parsed.data;
-}
 function parseValue(arg, argv, index) {
     const eqIndex = arg.indexOf("=");
     if (eqIndex !== -1) {
@@ -72,12 +55,11 @@ function parseValue(arg, argv, index) {
     }
     return { value: next, nextIndex: index + 1 };
 }
-export function parseCliArgs(argv, personas) {
+export function parseCliArgs(argv) {
     let help = false;
     let debug = false;
     let personaId;
     let reasoningOverride;
-    let riskLevel;
     let caffeinated = false;
     let noAgentContextFiles = false;
     for (let i = 0; i < argv.length; i++) {
@@ -103,18 +85,12 @@ export function parseCliArgs(argv, personas) {
             arg.startsWith("--persona=") ||
             arg.startsWith("-p=")) {
             const parsed = parseValue(arg, argv, i);
-            const persona = parsePersonaOption(parsed.value, personas);
+            const persona = parsePersonaOption(parsed.value);
             personaId = persona.personaId;
             reasoningOverride = persona.reasoningOverride;
             i = parsed.nextIndex;
             continue;
         }
-        if (arg === "--risk" || arg === "-r" || arg.startsWith("--risk=") || arg.startsWith("-r=")) {
-            const parsed = parseValue(arg, argv, i);
-            riskLevel = parseRiskOption(parsed.value);
-            i = parsed.nextIndex;
-            continue;
-        }
         if (arg.startsWith("-")) {
             throw new CliError(`unknown option: ${arg}`);
         }
@@ -125,7 +101,6 @@ export function parseCliArgs(argv, personas) {
         debug,
         personaId,
         reasoningOverride,
-        riskLevel,
         caffeinated,
         noAgentContextFiles,
     };
@@ -133,7 +108,6 @@ export function parseCliArgs(argv, personas) {
 export function printHelp(personas) {
     const personaList = personas.map((p) => p.id).join(", ");
     const reasoningList = REASONING_LEVELS.join(", ");
-    const riskList = RiskLevelSchema.options.join(", ");
     console.log([
         "tau - terminal chat",
         "",
@@ -153,12 +127,10 @@ export function printHelp(personas) {
         "",
         "options:",
         "  --help                        show this help and exit.",
-        "  --debug                       print debug info (personas, prompts, skills, system prompt) and exit.",
+        "  --debug                       print debug info and exit (TUI mode only).",
         `  --persona, -p <id>[:<level>]  start with a persona. available: ${personaList}.`,
         `                                optionally specify reasoning level. levels: ${reasoningList}.`,
         `                                if not specified, uses resolved config defaultPersona.`,
-        `  --risk, -r <level>            set initial model risk level. levels: ${riskList}.`,
-        `                                if not specified, uses resolved config defaultRisk (default: read-only).`,
         "  --caffeinated                 keep macOS awake during active assistant turns in TUI mode (no-op on Linux).",
         "  --no-agent-context-files      disable AGENTS.md injection into the system prompt.",
         "",
@@ -177,12 +149,8 @@ export function printHelp(personas) {
         "examples:",
         "  tau --persona gpt-5.4-chat:high",
         "  tau -p opus-4.8-coder",
-        "  tau --persona gpt-5.4-chat:medium --risk read-write",
-        "  tau -p gpt-5.4-coder:high -r read-write",
-        "  tau serve --host 0.0.0.0 --port 8787 --risk read-only",
         "  tau attach --new --cwd /repo ws://127.0.0.1:8787",
         "  tau attach ws://127.0.0.1:8787",
-        "  tau attach --new --cwd /repo -- ssh vps 'tau rpc --risk read-only'",
         "",
         "notes:",
         "  use `tau auth login codex` to authenticate ChatGPT subscription credentials.",
@@ -194,7 +162,6 @@ export function printHelp(personas) {
         "  /diff opens the local diff review tool and delegates review work to the session host.",
         "  you can switch persona during a session with /persona:<id>.",
         "  insert prompt templates with /prompt:<id>.",
-        "  you can change model risk level during a session with /risk:read-only or /risk:read-write.",
         "  if stdin is piped, its contents are sent as the first message automatically in TUI mode.",
         "  in RPC mode, stdin/stdout are reserved for protocol traffic.",
         "  reasoning only affects providers that support it.",

package/dist/core/cli.js.map

@@ -1 +1 @@
-{"version":3,"file":"cli.js","sourceRoot":"","sources":["../../src/core/cli.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC/D,OAAO,EAEL,gBAAgB,EAGhB,eAAe,GAChB,MAAM,YAAY,CAAC;AAYpB,MAAM,OAAO,QAAS,SAAQ,KAAK;IACjC,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,UAAU,CAAC;IACzB,CAAC;CACF;AAED,MAAM,UAAU,gBAAgB,CAAC,GAAW,EAAE,QAAmB;IAC/D,MAAM,OAAO,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;IAC3B,IAAI,CAAC,OAAO;QAAE,OAAO,SAAS,CAAC;IAC/B,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,OAAO,CAAC,EAAE,EAAE,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,kBAAkB,CAChC,GAAW,EACX,QAAmB;IAEnB,MAAM,MAAM,GAAG,qBAAqB,CAAC,GAAG,CAAC,CAAC;IAC1C,IAAI,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QACtC,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;IACxD,CAAC;IAED,MAAM,SAAS,GAAG,gBAAgB,CAAC,MAAM,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAC/D,OAAO;QACL,SAAS;QACT,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;KACpD,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CACzB,GAAW,EACX,QAAmB;IAEnB,MAAM,eAAe,GAAG,qBAAqB,CAAC,GAAG,CAAC,CAAC;IAEnD,IAAI,eAAe,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;QAClD,MAAM,IAAI,QAAQ,CAAC,gDAAgD,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,eAAe,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;QAClD,MAAM,OAAO,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC5C,MAAM,IAAI,QAAQ,CAChB,4BAA4B,eAAe,CAAC,YAAY,sBAAsB,OAAO,EAAE,CACxF,CAAC;IACJ,CAAC;IAED,IAAI,CAAC,eAAe,CAAC,SAAS,EAAE,CAAC;QAC/B,MAAM,IAAI,QAAQ,CAAC,iCAAiC,CAAC,CAAC;IACxD,CAAC;IAED,MAAM,SAAS,GAAG,gBAAgB,CAAC,eAAe,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IACxE,IAAI,CAAC,SAAS,EAAE,CAAC;QACf,MAAM,SAAS,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,IAAI,QAAQ,CAChB,oBAAoB,eAAe,CAAC,SAAS,0BAA0B,SAAS,EAAE,CACnF,CAAC;IACJ,CAAC;IAED,OAAO;QACL,SAAS;QACT,iBAAiB,EAAE,eAAe,CAAC,SAAS;KAC7C,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,GAAW;IAClC,MAAM,UAAU,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;IAC9B,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,MAAM,IAAI,QAAQ,CAAC,0BAA0B,CAAC,CAAC;IACjD,CAAC;IAED,MAAM,MAAM,GAAG,eAAe,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;IACrD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,OAAO,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnD,MAAM,IAAI,QAAQ,CAAC,uBAAuB,GAAG,sBAAsB,OAAO,EAAE,CAAC,CAAC;IAChF,CAAC;IAED,OAAO,MAAM,CAAC,IAAI,CAAC;AACrB,CAAC;AAED,SAAS,UAAU,CACjB,GAAW,EACX,IAAc,EACd,KAAa;IAEb,MAAM,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACjC,IAAI,OAAO,KAAK,CAAC,CAAC,EAAE,CAAC;QACnB,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;QACrC,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,QAAQ,CAAC,qBAAqB,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,CAAC,CAAC;QACnE,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;IACrC,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;IAC7B,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;QAClC,MAAM,IAAI,QAAQ,CAAC,qBAAqB,GAAG,EAAE,CAAC,CAAC;IACjD,CAAC;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,GAAG,CAAC,EAAE,CAAC;AAC/C,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,IAAc,EAAE,QAAmB;IAC9D,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,IAAI,KAAK,GAAG,KAAK,CAAC;IAClB,IAAI,SAA6B,CAAC;IAClC,IAAI,iBAA8C,CAAC;IACnD,IAAI,SAAgC,CAAC;IACrC,IAAI,WAAW,GAAG,KAAK,CAAC;IACxB,IAAI,mBAAmB,GAAG,KAAK,CAAC;IAEhC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,MAAM,GAAG,GAAG,IAAI,CAAC,CAAC,CAAE,CAAC;QAErB,IAAI,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC;YACrC,IAAI,GAAG,IAAI,CAAC;YACZ,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;YACtB,KAAK,GAAG,IAAI,CAAC;YACb,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,0BAA0B,EAAE,CAAC;YACvC,mBAAmB,GAAG,IAAI,CAAC;YAC3B,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,eAAe,EAAE,CAAC;YAC5B,WAAW,GAAG,IAAI,CAAC;YACnB,SAAS;QACX,CAAC;QAED,IACE,GAAG,KAAK,WAAW;YACnB,GAAG,KAAK,IAAI;YACZ,GAAG,CAAC,UAAU,CAAC,YAAY,CAAC;YAC5B,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,EACrB,CAAC;YACD,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;YACxC,MAAM,OAAO,GAAG,kBAAkB,CAAC,MAAM,CAAC,KAAK,EAAE,QAAQ,CAAC,CAAC;YAC3D,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC;YAC9B,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;YAC9C,CAAC,GAAG,MAAM,CAAC,SAAS,CAAC;YACrB,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,IAAI,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,IAAI,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;YAC3F,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;YACxC,SAAS,GAAG,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YAC1C,CAAC,GAAG,MAAM,CAAC,SAAS,CAAC;YACrB,SAAS;QACX,CAAC;QAED,IAAI,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YACxB,MAAM,IAAI,QAAQ,CAAC,mBAAmB,GAAG,EAAE,CAAC,CAAC;QAC/C,CAAC;QAED,MAAM,IAAI,QAAQ,CAAC,wBAAwB,GAAG,EAAE,CAAC,CAAC;IACpD,CAAC;IAED,OAAO;QACL,IAAI;QACJ,KAAK;QACL,SAAS;QACT,iBAAiB;QACjB,SAAS;QACT,WAAW;QACX,mBAAmB;KACpB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,QAAmB;IAC3C,MAAM,WAAW,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACzD,MAAM,aAAa,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAClD,MAAM,QAAQ,GAAG,eAAe,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEpD,OAAO,CAAC,GAAG,CACT;QACE,qBAAqB;QACrB,EAAE;QACF,QAAQ;QACR,iBAAiB;QACjB,qBAAqB;QACrB,uDAAuD;QACvD,8GAA8G;QAC9G,+FAA+F;QAC/F,yBAAyB;QACzB,uBAAuB;QACvB,yBAAyB;QACzB,yBAAyB;QACzB,sBAAsB;QACtB,qCAAqC;QACrC,0BAA0B;QAC1B,EAAE;QACF,UAAU;QACV,0DAA0D;QAC1D,uGAAuG;QACvG,oEAAoE,WAAW,GAAG;QAClF,+EAA+E,aAAa,GAAG;QAC/F,wFAAwF;QACxF,yEAAyE,QAAQ,GAAG;QACpF,0GAA0G;QAC1G,8GAA8G;QAC9G,qFAAqF;QACrF,EAAE;QACF,cAAc;QACd,uEAAuE;QACvE,mEAAmE;QACnE,mFAAmF;QACnF,gFAAgF;QAChF,uDAAuD;QACvD,qEAAqE;QACrE,oFAAoF;QACpF,6DAA6D;QAC7D,+DAA+D;QAC/D,yEAAyE;QACzE,EAAE;QACF,WAAW;QACX,mCAAmC;QACnC,yBAAyB;QACzB,uDAAuD;QACvD,2CAA2C;QAC3C,yDAAyD;QACzD,oDAAoD;QACpD,kCAAkC;QAClC,sEAAsE;QACtE,EAAE;QACF,QAAQ;QACR,gFAAgF;QAChF,qFAAqF;QACrF,uEAAuE;QACvE,4DAA4D;QAC5D,+EAA+E;QAC/E,+FAA+F;QAC/F,yFAAyF;QACzF,+DAA+D;QAC/D,8CAA8C;QAC9C,8FAA8F;QAC9F,4FAA4F;QAC5F,gEAAgE;QAChE,qDAAqD;KACtD,CAAC,IAAI,CAAC,IAAI,CAAC,CACb,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB;IAC/B,OAAO,CAAC,GAAG,CACT;QACE,QAAQ;QACR,0BAA0B;QAC1B,EAAE;QACF,uGAAuG;QACvG,0HAA0H;QAC1H,EAAE;QACF,WAAW;QACX,wBAAwB;KACzB,CAAC,IAAI,CAAC,IAAI,CAAC,CACb,CAAC;AACJ,CAAC"}
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../../src/core/cli.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC/D,OAAO,EAAgB,gBAAgB,EAAwB,MAAM,YAAY,CAAC;AAWlF,MAAM,OAAO,QAAS,SAAQ,KAAK;IACjC,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,UAAU,CAAC;IACzB,CAAC;CACF;AAED,MAAM,UAAU,gBAAgB,CAAC,GAAW,EAAE,QAAmB;IAC/D,MAAM,OAAO,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;IAC3B,IAAI,CAAC,OAAO;QAAE,OAAO,SAAS,CAAC;IAC/B,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,OAAO,CAAC,EAAE,EAAE,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,kBAAkB,CAChC,GAAW,EACX,QAAmB;IAEnB,MAAM,MAAM,GAAG,qBAAqB,CAAC,GAAG,CAAC,CAAC;IAC1C,IAAI,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;QACtC,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;IACxD,CAAC;IAED,MAAM,SAAS,GAAG,gBAAgB,CAAC,MAAM,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAC/D,OAAO;QACL,SAAS;QACT,SAAS,EAAE,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;KACpD,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CAAC,GAAW;IAIrC,MAAM,eAAe,GAAG,qBAAqB,CAAC,GAAG,CAAC,CAAC;IAEnD,IAAI,eAAe,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;QAClD,MAAM,IAAI,QAAQ,CAAC,gDAAgD,CAAC,CAAC;IACvE,CAAC;IAED,IAAI,eAAe,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;QAClD,MAAM,OAAO,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC5C,MAAM,IAAI,QAAQ,CAChB,4BAA4B,eAAe,CAAC,YAAY,sBAAsB,OAAO,EAAE,CACxF,CAAC;IACJ,CAAC;IAED,IAAI,CAAC,eAAe,CAAC,SAAS,EAAE,CAAC;QAC/B,MAAM,IAAI,QAAQ,CAAC,iCAAiC,CAAC,CAAC;IACxD,CAAC;IAED,OAAO;QACL,SAAS,EAAE,eAAe,CAAC,SAAS;QACpC,iBAAiB,EAAE,eAAe,CAAC,SAAS;KAC7C,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CACjB,GAAW,EACX,IAAc,EACd,KAAa;IAEb,MAAM,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACjC,IAAI,OAAO,KAAK,CAAC,CAAC,EAAE,CAAC;QACnB,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC;QACrC,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,QAAQ,CAAC,qBAAqB,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,CAAC,CAAC;QACnE,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;IACrC,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;IAC7B,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;QAClC,MAAM,IAAI,QAAQ,CAAC,qBAAqB,GAAG,EAAE,CAAC,CAAC;IACjD,CAAC;IACD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,GAAG,CAAC,EAAE,CAAC;AAC/C,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,IAAc;IACzC,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,IAAI,KAAK,GAAG,KAAK,CAAC;IAClB,IAAI,SAA6B,CAAC;IAClC,IAAI,iBAA8C,CAAC;IACnD,IAAI,WAAW,GAAG,KAAK,CAAC;IACxB,IAAI,mBAAmB,GAAG,KAAK,CAAC;IAEhC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,MAAM,GAAG,GAAG,IAAI,CAAC,CAAC,CAAE,CAAC;QAErB,IAAI,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC;YACrC,IAAI,GAAG,IAAI,CAAC;YACZ,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;YACtB,KAAK,GAAG,IAAI,CAAC;YACb,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,0BAA0B,EAAE,CAAC;YACvC,mBAAmB,GAAG,IAAI,CAAC;YAC3B,SAAS;QACX,CAAC;QAED,IAAI,GAAG,KAAK,eAAe,EAAE,CAAC;YAC5B,WAAW,GAAG,IAAI,CAAC;YACnB,SAAS;QACX,CAAC;QAED,IACE,GAAG,KAAK,WAAW;YACnB,GAAG,KAAK,IAAI;YACZ,GAAG,CAAC,UAAU,CAAC,YAAY,CAAC;YAC5B,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,EACrB,CAAC;YACD,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;YACxC,MAAM,OAAO,GAAG,kBAAkB,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;YACjD,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC;YAC9B,iBAAiB,GAAG,OAAO,CAAC,iBAAiB,CAAC;YAC9C,CAAC,GAAG,MAAM,CAAC,SAAS,CAAC;YACrB,SAAS;QACX,CAAC;QAED,IAAI,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YACxB,MAAM,IAAI,QAAQ,CAAC,mBAAmB,GAAG,EAAE,CAAC,CAAC;QAC/C,CAAC;QAED,MAAM,IAAI,QAAQ,CAAC,wBAAwB,GAAG,EAAE,CAAC,CAAC;IACpD,CAAC;IAED,OAAO;QACL,IAAI;QACJ,KAAK;QACL,SAAS;QACT,iBAAiB;QACjB,WAAW;QACX,mBAAmB;KACpB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,QAAmB;IAC3C,MAAM,WAAW,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACzD,MAAM,aAAa,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAElD,OAAO,CAAC,GAAG,CACT;QACE,qBAAqB;QACrB,EAAE;QACF,QAAQ;QACR,iBAAiB;QACjB,qBAAqB;QACrB,uDAAuD;QACvD,8GAA8G;QAC9G,+FAA+F;QAC/F,yBAAyB;QACzB,uBAAuB;QACvB,yBAAyB;QACzB,yBAAyB;QACzB,sBAAsB;QACtB,qCAAqC;QACrC,0BAA0B;QAC1B,EAAE;QACF,UAAU;QACV,0DAA0D;QAC1D,4EAA4E;QAC5E,oEAAoE,WAAW,GAAG;QAClF,+EAA+E,aAAa,GAAG;QAC/F,wFAAwF;QACxF,8GAA8G;QAC9G,qFAAqF;QACrF,EAAE;QACF,cAAc;QACd,uEAAuE;QACvE,mEAAmE;QACnE,mFAAmF;QACnF,gFAAgF;QAChF,uDAAuD;QACvD,qEAAqE;QACrE,oFAAoF;QACpF,6DAA6D;QAC7D,+DAA+D;QAC/D,yEAAyE;QACzE,EAAE;QACF,WAAW;QACX,mCAAmC;QACnC,yBAAyB;QACzB,oDAAoD;QACpD,kCAAkC;QAClC,EAAE;QACF,QAAQ;QACR,gFAAgF;QAChF,qFAAqF;QACrF,uEAAuE;QACvE,4DAA4D;QAC5D,+EAA+E;QAC/E,+FAA+F;QAC/F,yFAAyF;QACzF,+DAA+D;QAC/D,8CAA8C;QAC9C,4FAA4F;QAC5F,gEAAgE;QAChE,qDAAqD;KACtD,CAAC,IAAI,CAAC,IAAI,CAAC,CACb,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB;IAC/B,OAAO,CAAC,GAAG,CACT;QACE,QAAQ;QACR,0BAA0B;QAC1B,EAAE;QACF,uGAAuG;QACvG,0HAA0H;QAC1H,EAAE;QACF,WAAW;QACX,wBAAwB;KACzB,CAAC,IAAI,CAAC,IAAI,CAAC,CACb,CAAC;AACJ,CAAC"}

package/dist/core/commands/index.js

@@ -1,2 +1,2 @@
-export { CommandRegistry, createCommandRegistry, getRiskLevelAutocompleteOptions, getRiskLevelDescription, } from "./registry.js";
+export { CommandRegistry, createCommandRegistry, } from "./registry.js";
 //# sourceMappingURL=index.js.map

package/dist/core/commands/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/core/commands/index.ts"],"names":[],"mappings":"AAQA,OAAO,EACL,eAAe,EACf,qBAAqB,EACrB,+BAA+B,EAC/B,uBAAuB,GACxB,MAAM,eAAe,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/core/commands/index.ts"],"names":[],"mappings":"AAQA,OAAO,EACL,eAAe,EACf,qBAAqB,GACtB,MAAM,eAAe,CAAC"}

package/dist/core/commands/registry.js

@@ -1,25 +1,5 @@
 import { dirname } from "node:path";
-import { RiskLevelSchema } from "../types.js";
 import { formatPathForDisplay } from "../utils/format.js";
-const DEFAULT_RISK_LEVELS = ["read-only", "read-write"];
-const RISK_LEVEL_HELP_DESCRIPTIONS = {
-    "read-only": "allow read-only tool calls",
-    "read-write": "allow all tools",
-};
-export function getRiskLevelDescription(level) {
-    switch (level) {
-        case "read-only":
-            return "read-only tools";
-        case "read-write":
-            return "all tools";
-    }
-}
-export function getRiskLevelAutocompleteOptions(allowed) {
-    return allowed.map((level) => ({
-        id: level,
-        description: RISK_LEVEL_HELP_DESCRIPTIONS[level],
-    }));
-}
 function formatSkillPath(fullPath, formatPath) {
     return formatPath(dirname(dirname(fullPath)));
 }
@@ -66,7 +46,7 @@ export class CommandRegistry {
     }
     buildHelpText(options = {}) {
         const lines = [];
-        const { agentsFiles, skills, riskLevels, themes } = options;
+        const { agentsFiles, skills, themes } = options;
         const formatPath = options.formatPath ?? formatPathForDisplay;
         if (agentsFiles && agentsFiles.length > 0) {
             lines.push("context:");
@@ -88,26 +68,12 @@ export class CommandRegistry {
         }
         const hasThemes = (themes?.length ?? 0) > 0;
         const commands = this.list().filter((command) => command.argument !== "theme" || hasThemes);
-        const baseCommands = commands.filter((command) => command.section === "base");
-        const trailingCommands = commands.filter((command) => command.section === "trailing");
-        const riskCommand = commands.find((command) => command.section === "risk");
-        const commandEntries = [];
-        baseCommands.forEach((command) => {
-            commandEntries.push([command.usage, command.description]);
-        });
-        if (riskCommand) {
-            const allowed = riskLevels ?? DEFAULT_RISK_LEVELS;
-            allowed.forEach((level) => {
-                const description = RISK_LEVEL_HELP_DESCRIPTIONS[level];
-                commandEntries.push([`/risk:${level}`, description]);
-            });
-        }
-        trailingCommands.forEach((command) => {
-            commandEntries.push([command.usage, command.description]);
-        });
+        const commandEntries = commands.map((command) => [
+            command.usage,
+            command.description,
+        ]);
         const keyEntries = [
             ["shift+tab", "cycle reasoning effort"],
-            ["ctrl+r", "cycle risk level"],
             ["ctrl+p", "cycle personality"],
             ["ctrl+t", "toggle thoughts visibility"],
             ["ctrl+o", "toggle compact tool UI"],
@@ -140,7 +106,6 @@ export function createCommandRegistry() {
         description: "show this help",
         autocompleteDescription: "show help",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/help")
@@ -155,7 +120,6 @@ export function createCommandRegistry() {
         description: "exit tau",
         autocompleteDescription: "exit tau",
         argument: "none",
-        section: "base",
         allowDuringStreaming: true,
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
@@ -171,7 +135,6 @@ export function createCommandRegistry() {
         description: "new session",
         autocompleteDescription: "new session",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/new")
@@ -186,7 +149,6 @@ export function createCommandRegistry() {
         description: "rewind context to an earlier user message",
         autocompleteDescription: "rewind context to a selected user message",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/rewind")
@@ -201,7 +163,6 @@ export function createCommandRegistry() {
         description: "open the local diff review tool for the session diff",
         autocompleteDescription: "open diff review for git diff args",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/diff")
@@ -216,7 +177,6 @@ export function createCommandRegistry() {
         description: "summarize and start new session",
         autocompleteDescription: "compact history to a summary",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/compact:summary-only")
@@ -231,7 +191,6 @@ export function createCommandRegistry() {
         description: "summarize and include previous last assistant message",
         autocompleteDescription: "compact history, keep last assistant message",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/compact:summary-and-last")
@@ -246,7 +205,6 @@ export function createCommandRegistry() {
         description: "prune earliest tool results from context",
         autocompleteDescription: "prune earliest tool results",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/prune:earliest")
@@ -261,7 +219,6 @@ export function createCommandRegistry() {
         description: "prune largest tool results from context",
         autocompleteDescription: "prune largest tool results",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/prune:largest")
@@ -276,7 +233,6 @@ export function createCommandRegistry() {
         description: "prune tool results using model selection",
         autocompleteDescription: "prune tool results with model selection",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/prune:smart")
@@ -291,7 +247,6 @@ export function createCommandRegistry() {
         description: "reload prompts, skills, themes, and AGENTS.md",
         autocompleteDescription: "reload prompts, skills, themes, and AGENTS.md",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/reload")
@@ -306,7 +261,6 @@ export function createCommandRegistry() {
         description: "start voice recording and transcription",
         autocompleteDescription: "start voice recording",
         argument: "none",
-        section: "base",
         allowDuringStreaming: true,
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
@@ -322,7 +276,6 @@ export function createCommandRegistry() {
         description: "speak the last assistant message aloud",
         autocompleteDescription: "speak the last assistant message",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/speak")
@@ -337,7 +290,6 @@ export function createCommandRegistry() {
         description: "copy last assistant message",
         autocompleteDescription: "copy last assistant message",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/copy:text")
@@ -352,7 +304,6 @@ export function createCommandRegistry() {
         description: "copy code blocks from last assistant message",
         autocompleteDescription: "copy code blocks from last assistant message",
         argument: "none",
-        section: "base",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             if (command !== "/copy:code")
@@ -361,30 +312,11 @@ export function createCommandRegistry() {
         },
         run: (ctx) => ctx.copyCode(),
     });
-    registry.register({
-        id: "risk",
-        usage: "/risk:<level>",
-        description: "set risk level",
-        argument: "risk",
-        section: "risk",
-        parse: (raw) => {
-            const { command, extra } = splitCommandInput(raw);
-            const match = command.match(/^\/risk:(read-only|read-write)$/i);
-            if (!match)
-                return null;
-            const parsed = RiskLevelSchema.safeParse(match[1]?.toLowerCase());
-            if (!parsed.success)
-                return null;
-            return { type: "risk", level: parsed.data, extra };
-        },
-        run: (ctx, command) => ctx.risk(command.level),
-    });
     registry.register({
         id: "persona",
         usage: "/persona:<id>",
         description: "switch persona",
         argument: "persona",
-        section: "trailing",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             const match = command.match(/^\/persona:(.+)$/i);
@@ -400,7 +332,6 @@ export function createCommandRegistry() {
         usage: "/prompt:<id>",
         description: "insert prompt template",
         argument: "prompt",
-        section: "trailing",
         allowDuringStreaming: true,
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
@@ -417,7 +348,6 @@ export function createCommandRegistry() {
         usage: "/theme:<id>",
         description: "switch theme",
         argument: "theme",
-        section: "trailing",
         parse: (raw) => {
             const { command, extra } = splitCommandInput(raw);
             const match = command.match(/^\/theme:(.+)$/i);
@@ -433,7 +363,6 @@ export function createCommandRegistry() {
         usage: "",
         description: "",
         argument: "none",
-        section: "base",
         hidden: true,
         parse: () => null,
         run: (ctx, command) => ctx.unknown(command.raw),