Merge branch 'main' of github.com:NousResearch/hermes-agent into bb/gui

run_agent.py

@@ -3465,6 +3465,15 @@ class AIAgent:
             return True, True
         if (is_openrouter or is_nous_portal) and is_claude:
             return True, False
+        # Nous Portal Qwen (e.g. qwen3.6-plus) takes the same envelope-layout
+        # cache_control path as Portal Claude. Portal proxies to OpenRouter
+        # and the upstream Qwen route accepts cache_control markers; without
+        # this branch the alibaba-family check below only matches
+        # provider=opencode/alibaba and Portal traffic falls through to
+        # (False, False), serving 0% cache hits and re-billing the full
+        # prompt on every turn.
+        if is_nous_portal and "qwen" in model_lower:
+            return True, False
         if is_anthropic_wire and is_claude:
             # Third-party Anthropic-compatible gateway.
             return True, True
@@ -3540,7 +3549,19 @@ class AIAgent:
         eff_api_mode = api_mode if api_mode is not None else (self.api_mode or "")
         eff_model = (model if model is not None else self.model) or ""
 
-        if "claude" not in eff_model.lower():
+        model_lower = eff_model.lower()
+        is_claude = "claude" in model_lower
+        is_nous_portal = "nousresearch" in eff_base_url.lower()
+
+        # Nous Portal: Claude AND Qwen both get long-lived caching.
+        # Portal proxies to OpenRouter with identical cache_control
+        # semantics; any model on Portal that accepts envelope-layout
+        # markers via _anthropic_prompt_cache_policy also benefits from
+        # the documented 1h cross-session TTL.
+        if is_nous_portal and (is_claude or "qwen" in model_lower):
+            return True
+
+        if not is_claude:
             return False
 
         # Native Anthropic + Anthropic OAuth subscription
@@ -3554,7 +3575,7 @@ class AIAgent:
 
         # Nous Portal — front-ends OpenRouter behind the scenes; identical
         # wire format and cache_control semantics.
-        if "nousresearch" in eff_base_url.lower():
+        if is_nous_portal:
             return True
 
         return False

tests/run_agent/test_anthropic_prompt_cache_policy.py

@@ -257,6 +257,40 @@ class TestQwenAlibabaFamily:
         )
         assert agent._anthropic_prompt_cache_policy() == (False, False)
 
+    def test_qwen_on_nous_portal_caches_with_envelope_layout(self):
+        # Nous Portal Qwen takes the same envelope-layout cache_control
+        # path as Portal Claude. Without this, Portal-routed qwen3.6-plus
+        # falls through to the alibaba-family check (which only matches
+        # provider=opencode/alibaba) and serves 0% cache hits.
+        agent = _make_agent(
+            provider="nous",
+            base_url="https://inference-api.nousresearch.com/v1",
+            api_mode="chat_completions",
+            model="qwen3.6-plus",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (True, False)
+
+    def test_qwen_vendored_slug_on_nous_portal_caches(self):
+        # Same path but with the vendored slug form Portal sometimes uses.
+        agent = _make_agent(
+            provider="nous",
+            base_url="https://inference-api.nousresearch.com/v1",
+            api_mode="chat_completions",
+            model="qwen/qwen3.6-plus",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (True, False)
+
+    def test_non_qwen_non_claude_on_nous_portal_does_not_cache(self):
+        # Portal scope is narrow: Claude OR Qwen only. Other models
+        # routed through Portal keep their existing fall-through behavior.
+        agent = _make_agent(
+            provider="nous",
+            base_url="https://inference-api.nousresearch.com/v1",
+            api_mode="chat_completions",
+            model="openai/gpt-5.4",
+        )
+        assert agent._anthropic_prompt_cache_policy() == (False, False)
+
 
 class TestExplicitOverrides:
     """Policy accepts keyword overrides for switch_model / fallback activation."""
@@ -338,6 +372,37 @@ class TestSupportsLongLivedAnthropicCache:
         )
         assert agent._supports_long_lived_anthropic_cache() is True
 
+    def test_nous_portal_qwen_supported(self):
+        # Portal Qwen rides the same OpenRouter-equivalent transport as
+        # Portal Claude; long-lived (1h cross-session) cache_control
+        # markers apply identically.
+        agent = _make_agent(
+            provider="nous",
+            base_url="https://inference-api.nousresearch.com/v1",
+            api_mode="chat_completions",
+            model="qwen3.6-plus",
+        )
+        assert agent._supports_long_lived_anthropic_cache() is True
+
+    def test_nous_portal_qwen_vendored_slug_supported(self):
+        agent = _make_agent(
+            provider="nous",
+            base_url="https://inference-api.nousresearch.com/v1",
+            api_mode="chat_completions",
+            model="qwen/qwen3.6-plus",
+        )
+        assert agent._supports_long_lived_anthropic_cache() is True
+
+    def test_nous_portal_non_claude_non_qwen_rejected(self):
+        # Portal long-lived cache scope mirrors policy: Claude or Qwen only.
+        agent = _make_agent(
+            provider="nous",
+            base_url="https://inference-api.nousresearch.com/v1",
+            api_mode="chat_completions",
+            model="openai/gpt-5.4",
+        )
+        assert agent._supports_long_lived_anthropic_cache() is False
+
     def test_openrouter_non_claude_rejected(self):
         agent = _make_agent(
             provider="openrouter",

ui-tui/packages/hermes-ink/src/ink/termio/osc.test.ts

@@ -1,6 +1,8 @@
 import { describe, expect, it } from 'vitest'
 
-import { shouldEmitClipboardSequence } from './osc.js'
+import { env, supportsOsc52Clipboard } from '../../utils/env.js'
+
+import { shouldEmitClipboardSequence, shouldUseNativeClipboard } from './osc.js'
 
 describe('shouldEmitClipboardSequence', () => {
   it('suppresses local multiplexer clipboard OSC by default', () => {
@@ -49,3 +51,141 @@ describe('shouldEmitClipboardSequence', () => {
     ).toBe(false)
   })
 })
+
+describe('supportsOsc52Clipboard', () => {
+  // Terminals known to correctly implement OSC 52. On these, setClipboard()
+  // skips the native-tool safety net (wl-copy/xclip/pbcopy) to avoid racing
+  // the terminal's own clipboard write. Values must match what
+  // detectTerminal() in utils/env.ts returns — TERM=xterm-ghostty normalises
+  // to 'ghostty', TERM_PROGRAM=WezTerm stays 'WezTerm', etc.
+  it.each(['ghostty', 'kitty', 'WezTerm', 'windows-terminal', 'vscode'])(
+    'returns true for allowlisted terminal %s',
+    terminal => {
+      expect(supportsOsc52Clipboard(terminal)).toBe(true)
+    }
+  )
+
+  // Intentionally conservative — iTerm2 disables OSC 52 by default; Alacritty
+  // and GNOME Terminal detection is unreliable; xterm/Terminal.app lack
+  // reliable OSC 52. These keep the existing native-safety-net behaviour.
+  it.each(['iTerm.app', 'alacritty', 'Apple_Terminal', 'xterm', 'tmux', 'screen', 'cursor', 'WarpTerminal', ''])(
+    'returns false for non-allowlisted terminal %s',
+    terminal => {
+      expect(supportsOsc52Clipboard(terminal)).toBe(false)
+    }
+  )
+
+  it('returns false when terminal is null (detection failed)', () => {
+    expect(supportsOsc52Clipboard(null)).toBe(false)
+  })
+
+  it('defaults to the module-level detected terminal when no argument is passed', () => {
+    // With no argument, uses env.terminal detected at module load. We don't
+    // know what that is in CI, but the call must return a boolean (not throw)
+    // and the result must match calling with env.terminal explicitly.
+    expect(typeof supportsOsc52Clipboard()).toBe('boolean')
+    expect(supportsOsc52Clipboard()).toBe(supportsOsc52Clipboard(env.terminal))
+  })
+})
+
+// shouldUseNativeClipboard() encodes the gating logic that setClipboard()
+// uses to decide whether to fire copyNative(). Testing it directly (rather
+// than mocking copyNative inside setClipboard) matches the package's
+// existing style — tests pass env/terminal as arguments instead of using
+// vi.mock — and gives broader coverage of the env x terminal matrix.
+describe('shouldUseNativeClipboard', () => {
+  it('returns false over SSH (native would write to remote clipboard)', () => {
+    // Over SSH the user's terminal is on the local end of the pty;
+    // pbcopy/wl-copy/xclip on the remote machine would write to the wrong
+    // clipboard. OSC 52 is the right path. Existing behaviour, preserved.
+    expect(shouldUseNativeClipboard({ SSH_CONNECTION: '1' } as NodeJS.ProcessEnv, 'xterm')).toBe(false)
+    expect(shouldUseNativeClipboard({ SSH_CONNECTION: '1' } as NodeJS.ProcessEnv, 'ghostty')).toBe(false)
+    expect(shouldUseNativeClipboard({ SSH_CONNECTION: '1' } as NodeJS.ProcessEnv, null)).toBe(false)
+  })
+
+  it('returns true on plain local terminals (existing behaviour)', () => {
+    // Non-allowlisted terminals — xterm, GNOME Terminal, Apple_Terminal,
+    // alacritty (detection unreliable), iTerm2 (OSC 52 off by default).
+    // These keep the native safety net firing. This is the bulk of the
+    // existing user base; behaviour must not regress.
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'xterm')).toBe(true)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'iTerm.app')).toBe(true)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'Apple_Terminal')).toBe(true)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'alacritty')).toBe(true)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, null)).toBe(true)
+  })
+
+  it('returns false on allowlisted local terminals (the race-fix case)', () => {
+    // Ghostty / kitty / WezTerm / Windows Terminal / VS Code — OSC 52
+    // alone is reliable, native fallback racing it can corrupt the
+    // clipboard (the wl-copy on Wayland symptom this PR fixes).
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'ghostty')).toBe(false)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'kitty')).toBe(false)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'WezTerm')).toBe(false)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'windows-terminal')).toBe(false)
+    expect(shouldUseNativeClipboard({} as NodeJS.ProcessEnv, 'vscode')).toBe(false)
+  })
+
+  it('returns true inside tmux even on allowlisted outer terminal', () => {
+    // detectTerminal() prefers TERM_PROGRAM over TMUX, so a tmux session
+    // inside Ghostty reports terminal='ghostty'. But setClipboard() goes
+    // through tmux load-buffer there, not raw OSC 52 — the wl-copy race
+    // doesn't apply. Native is still useful since tmux's outer-terminal
+    // forwarding depends on `set -g set-clipboard` + `allow-passthrough`.
+    expect(shouldUseNativeClipboard({ TMUX: '/tmp/t,1,0' } as NodeJS.ProcessEnv, 'ghostty')).toBe(true)
+    expect(shouldUseNativeClipboard({ TMUX: '/tmp/t,1,0' } as NodeJS.ProcessEnv, 'kitty')).toBe(true)
+    expect(shouldUseNativeClipboard({ TMUX: '/tmp/t,1,0' } as NodeJS.ProcessEnv, 'WezTerm')).toBe(true)
+    expect(shouldUseNativeClipboard({ TMUX: '/tmp/t,1,0' } as NodeJS.ProcessEnv, 'vscode')).toBe(true)
+  })
+
+  it('returns true inside GNU screen even on allowlisted outer terminal', () => {
+    // Same reasoning as TMUX — STY indicates we're inside screen, which
+    // has its own escape-sequence handling and we don't emit raw OSC 52.
+    expect(shouldUseNativeClipboard({ STY: '1234.pts-0.host' } as NodeJS.ProcessEnv, 'ghostty')).toBe(true)
+    expect(shouldUseNativeClipboard({ STY: '1234.pts-0.host' } as NodeJS.ProcessEnv, 'kitty')).toBe(true)
+  })
+
+  it('returns true when OSC 52 emission is disabled via HERMES_TUI_FORCE_OSC52=0', () => {
+    // If we suppress OSC 52 (user override) AND skip native, the clipboard
+    // write becomes a no-op. So when OSC 52 is off, native is the only
+    // remaining path — keep it on regardless of terminal allowlist.
+    expect(shouldUseNativeClipboard({ HERMES_TUI_FORCE_OSC52: '0' } as NodeJS.ProcessEnv, 'ghostty')).toBe(true)
+    expect(shouldUseNativeClipboard({ HERMES_TUI_FORCE_OSC52: '0' } as NodeJS.ProcessEnv, 'kitty')).toBe(true)
+    expect(shouldUseNativeClipboard({ HERMES_TUI_CLIPBOARD_OSC52: '0' } as NodeJS.ProcessEnv, 'WezTerm')).toBe(true)
+    expect(shouldUseNativeClipboard({ HERMES_TUI_COPY_OSC52: 'no' } as NodeJS.ProcessEnv, 'vscode')).toBe(true)
+  })
+
+  it('returns true under TMUX even with HERMES_TUI_FORCE_OSC52=1 on an allowlisted terminal (tmux load-buffer path)', () => {
+    // FORCE_OSC52=1 is the user explicitly opting INTO OSC 52 (e.g. they
+    // have tmux set up for passthrough). On an allowlisted terminal the
+    // race-avoidance still applies.
+    expect(
+      shouldUseNativeClipboard({ HERMES_TUI_FORCE_OSC52: '1', TMUX: '/tmp/t,1,0' } as NodeJS.ProcessEnv, 'ghostty')
+      // TMUX guard wins — native still fires because we're going through
+      // tmux load-buffer, not raw OSC 52 to the terminal.
+    ).toBe(true)
+  })
+
+  it('SSH_CONNECTION takes precedence over allowlisted terminal', () => {
+    // Even on Ghostty, if we're SSH'd in we shouldn't run pbcopy on the
+    // remote machine — the user's clipboard is on the other end.
+    expect(shouldUseNativeClipboard({ SSH_CONNECTION: '1' } as NodeJS.ProcessEnv, 'ghostty')).toBe(false)
+  })
+
+  it('SSH_CONNECTION takes precedence over TMUX', () => {
+    // Combined: SSH'd in and inside tmux on the remote. SSH_CONNECTION
+    // gate fires first, native stays off (we use OSC 52 to reach the
+    // local terminal).
+    expect(shouldUseNativeClipboard({ SSH_CONNECTION: '1', TMUX: '/tmp/t,1,0' } as NodeJS.ProcessEnv, 'xterm')).toBe(
+      false
+    )
+  })
+
+  it('defaults env to process.env and terminal to the module-detected terminal when no args passed', () => {
+    // Smoke test: no args is a valid call shape for the convenience seam.
+    // shouldUseNativeClipboard() defaults `terminal` to envModule.terminal
+    // (the module-level detected terminal), not null. Returns a boolean
+    // without throwing.
+    expect(typeof shouldUseNativeClipboard()).toBe('boolean')
+  })
+})

ui-tui/packages/hermes-ink/src/ink/termio/osc.ts

@@ -4,7 +4,7 @@
 
 import { Buffer } from 'buffer'
 
-import { env } from '../../utils/env.js'
+import { env as envModule, supportsOsc52Clipboard } from '../../utils/env.js'
 import { execFileNoThrow } from '../../utils/execFileNoThrow.js'
 
 import { BEL, ESC, ESC_TYPE, SEP } from './ansi.js'
@@ -20,7 +20,7 @@ export const ST = ESC + '\\'
 /** Generate an OSC sequence: ESC ] p1;p2;...;pN <terminator>
  * Uses ST terminator for Kitty (avoids beeps), BEL for others */
 export function osc(...parts: (string | number)[]): string {
-  const terminator = env.terminal === 'kitty' ? ST : BEL
+  const terminator = envModule.terminal === 'kitty' ? ST : BEL
 
   return `${OSC_PREFIX}${parts.join(SEP)}${terminator}`
 }
@@ -102,6 +102,77 @@ export function shouldEmitClipboardSequence(env: NodeJS.ProcessEnv = process.env
   return !!env['SSH_CONNECTION'] || (!env['TMUX'] && !env['STY'])
 }
 
+/**
+ * Decide whether setClipboard() should also fire the native clipboard tool
+ * (pbcopy / wl-copy / xclip / xsel / clip.exe) as a safety net alongside
+ * OSC 52 / tmux load-buffer.
+ *
+ * The default is "yes, native fires" — it's the historical safety net for
+ * terminals where OSC 52 may not work (iTerm2 disables OSC 52 by default,
+ * Apple_Terminal / GNOME Terminal / xterm coverage is patchy). The two
+ * cases where we suppress it:
+ *
+ *  1. SSH session: native tools would write to the *remote* machine's
+ *     clipboard. OSC 52 (which travels back over the pty to the user's
+ *     local terminal) is the right path. Existing behaviour.
+ *
+ *  2. Allowlisted OSC-52-capable terminal AND we're actually going to
+ *     emit an OSC 52 sequence AND we're not inside tmux/screen. On these
+ *     terminals (Ghostty / kitty / WezTerm / Windows Terminal / VS Code)
+ *     the OSC 52 write is reliable on its own, and racing it with a
+ *     native tool is destructive — wl-copy on Wayland in particular
+ *     wipes the clipboard during its existence-probe and forks a daemon
+ *     that races the terminal's own write (~30% empty-clipboard rate
+ *     reported on Ghostty + Wayland; symptom: ctrl+shift+c works on the
+ *     3rd attempt).
+ *
+ *     The TMUX/STY guard is important: detectTerminal() in utils/env.ts
+ *     prefers TERM_PROGRAM over TMUX, so a tmux session inside Ghostty
+ *     reports terminal='ghostty'. But inside tmux setClipboard() doesn't
+ *     emit raw OSC 52 — it goes through tmux load-buffer (which loads
+ *     the tmux paste buffer and, with -w, asks tmux to forward an OSC 52
+ *     to the OUTER terminal via its own emission path). The native
+ *     safety net is still useful there because tmux load-buffer's
+ *     outer-terminal forwarding depends on `set -g set-clipboard` and
+ *     `allow-passthrough`, which many users don't have configured.
+ *
+ *     The OSC-52-will-emit guard matters too: if the user has set
+ *     HERMES_TUI_FORCE_OSC52=0, no OSC 52 sequence will be written. If
+ *     we ALSO skip native, the clipboard write becomes a no-op. So skip
+ *     native only when OSC 52 will actually carry the data.
+ */
+export function shouldUseNativeClipboard(
+  env: NodeJS.ProcessEnv = process.env,
+  terminal: string | null = envModule.terminal
+): boolean {
+  // Over SSH the native tools would write to the wrong machine's clipboard.
+  if (env.SSH_CONNECTION) {
+    return false
+  }
+
+  // Inside tmux/screen, OSC 52 is normally suppressed and we rely on
+  // tmux load-buffer instead — so the wl-copy/OSC-52 race usually doesn't
+  // apply. Even when HERMES_TUI_FORCE_OSC52=1 forces a tmux-passthrough
+  // OSC 52 emission, we keep native enabled as a safety net: tmux's
+  // outer-terminal forwarding depends on `allow-passthrough` in the
+  // user's tmux config, so a forced OSC 52 may silently never reach the
+  // host terminal. Native (pbcopy/wl-copy/xclip) covers that gap.
+  if (env.TMUX || env.STY) {
+    return true
+  }
+
+  // If OSC 52 won't actually emit (user override or env state), the
+  // native tool is the only path left — keep it on.
+  if (!shouldEmitClipboardSequence(env)) {
+    return true
+  }
+
+  // OSC 52 is going to emit AND the terminal is in the allowlist of
+  // terminals where OSC 52 alone is reliable: skip native to avoid the
+  // wl-copy race documented above.
+  return !supportsOsc52Clipboard(terminal)
+}
+
 /**
  * Wrap a payload in tmux's DCS passthrough: ESC P tmux ; <payload> ESC \
  * tmux forwards the payload to the outer terminal, bypassing its own parser.
@@ -193,11 +264,27 @@ export async function setClipboard(text: string): Promise<ClipboardResult> {
   // AFTER awaiting tmux load-buffer, adding ~50-100ms of subprocess latency
   // before pbcopy even started — fast cmd+tab → paste would beat it
   // (https://anthropic.slack.com/archives/C07VBSHV7EV/p1773943921788829).
-  // Gated on SSH_CONNECTION (not SSH_TTY) since tmux panes inherit SSH_TTY
-  // forever but SSH_CONNECTION is in tmux's default update-environment and
-  // clears on local attach. Fire-and-forget, but `copyNativeAttempted`
-  // tells us whether ANY native path will be tried on this platform.
-  const nativeAttempted = !process.env['SSH_CONNECTION'] && copyNative(text)
+  // Skipped entirely on terminals with first-class OSC 52 support (see
+  // `shouldUseNativeClipboard()` above): running wl-copy/xclip/pbcopy in
+  // parallel with OSC 52 on those terminals can corrupt the clipboard.
+  // wl-copy on Wayland is the worst offender — `probeLinuxCopy()` runs it
+  // with empty stdin to check if the binary exists (which destructively
+  // wipes the clipboard), and the subsequent real invocation forks a
+  // background daemon that races the terminal's own OSC 52 write plus its
+  // own prior daemon's SIGTERM. On Ghostty + Wayland this produced a ~30%
+  // clipboard-empty rate (symptom: user had to press ctrl+shift+c three
+  // times before the selection landed). Native still fires inside
+  // tmux/screen — we primarily rely on tmux load-buffer there rather
+  // than raw OSC 52, so the wl-copy race usually doesn't apply, and
+  // native is kept as a safety net because tmux passthrough forwarding
+  // depends on the user's `allow-passthrough` config (note: when
+  // HERMES_TUI_FORCE_OSC52=1 we DO additionally emit a tmux-passthrough
+  // OSC 52, but it can be silently dropped without that setting).
+  // Native also fires when the user has disabled OSC 52 emission via
+  // HERMES_TUI_FORCE_OSC52=0 (otherwise the clipboard write becomes a
+  // complete no-op). Fire-and-forget, but `nativeAttempted` tells us
+  // whether ANY native path will be tried.
+  const nativeAttempted = shouldUseNativeClipboard(process.env, envModule.terminal) && copyNative(text)
 
   const tmuxBufferLoaded = await tmuxLoadBuffer(text)
 

ui-tui/packages/hermes-ink/src/utils/env.ts

@@ -39,3 +39,28 @@ function detectTerminal(): TerminalName {
 export const env = {
   terminal: detectTerminal()
 }
+
+// Terminals known to correctly implement OSC 52 clipboard writes
+// (ESC ] 52 ; c ; <b64> BEL/ST — osc() in ink/termio/osc.ts emits BEL
+// for most terminals and ST for kitty). When detected, setClipboard() skips the
+// native-tool safety net entirely — running wl-copy/xclip/pbcopy in
+// parallel with OSC 52 races the terminal's own clipboard write and can
+// corrupt it (e.g. wl-copy on Wayland holds the selection in a background
+// daemon; stacking two writes within ~30ms triggers a SIGTERM race).
+// Intentionally conservative: terminals with known flaky or disabled-by-
+// default OSC 52 (iTerm2 disables OSC 52 by default; Alacritty detection
+// is unreliable) are not on this list. Users on those terminals keep the
+// existing behaviour (native safety net fires alongside OSC 52).
+//
+// Lives here in utils/env.ts (rather than ink/terminal.ts) so that
+// ink/termio/osc.ts can import it without creating a circular dependency:
+// ink/terminal.ts already imports `link` from ink/termio/osc.ts.
+const OSC52_CAPABLE_TERMINALS = ['ghostty', 'kitty', 'WezTerm', 'windows-terminal', 'vscode']
+
+/** True if this terminal is known to correctly handle OSC 52 clipboard
+ *  writes, so setClipboard() can skip the native-tool safety net.
+ *  Accepts an optional terminal name for testability; defaults to the
+ *  module-level `env.terminal` detected at startup. */
+export function supportsOsc52Clipboard(terminal: string | null = env.terminal): boolean {
+  return OSC52_CAPABLE_TERMINALS.includes(terminal ?? '')
+}