@bitkyc08/opencodex 2.7.2 → 2.7.3

package/gui/dist/assets/{index-DGFbiUtS.js → index-DT4C-vKW.js}

@@ -37,4 +37,4 @@ v2: 多线程代理（spawn_agent）。所有模型使用 v2 协作界面。
   -d '{
     "model": "gpt-5.4",
     "input": "Hello, world!"
-  }'`})]})]})}var Ft=`opencodex-api-token`,It=!1,Lt=null;function Rt(e){try{let t=e instanceof Request?e.url:String(e);return new URL(t,window.location.href).pathname}catch{return null}}function zt(e){let t=Rt(e);return!!t&&(t.startsWith(`/api/`)||t.startsWith(`/v1/`))}function Bt(){try{return sessionStorage.getItem(Ft)?.trim()||null}catch{return null}}function Vt(e){try{sessionStorage.setItem(Ft,e)}catch{}}function Ht(){try{sessionStorage.removeItem(Ft)}catch{}}function Ut(e,t,n){let r=new Headers(t?.headers??(e instanceof Request?e.headers:void 0));return r.set(`X-OpenCodex-API-Key`,n),e instanceof Request?[new Request(e,{headers:r}),t?{...t,headers:r}:void 0]:[e,{...t,headers:r}]}async function Wt(){return Lt||(Lt=Promise.resolve().then(()=>window.prompt(`OpenCodex API token`)?.trim()||null).finally(()=>{Lt=null}),Lt)}function Gt(){if(It)return;It=!0;let e=window.fetch.bind(window);window.fetch=async(t,n)=>{if(!zt(t))return e(t,n);let r=Bt(),[i,a]=r?Ut(t,n,r):[t,n],o=await e(i,a);if(o.status!==401)return o;r&&Ht();let s=await Wt();if(!s)return o;Vt(s);let[c,l]=Ut(t,n,s),u=await e(c,l);return u.status===401&&Ht(),u}}Gt();var Kt=new Set([`dashboard`,`providers`,`models`,`subagents`,`logs`,`debug`,`usage`,`codex-auth`,`api`]);function qt(){let e=location.hash.replace(/^#\/?/,``);return Kt.has(e)?e:`dashboard`}var Jt=``,Yt=`ocx-theme`,Xt=[{id:`dashboard`,tkey:`nav.dashboard`,Icon:w},{id:`providers`,tkey:`nav.providers`,Icon:T},{id:`models`,tkey:`nav.models`,Icon:E},{id:`subagents`,tkey:`nav.subagents`,Icon:ee},{id:`logs`,tkey:`nav.logs`,Icon:D},{id:`debug`,tkey:`nav.debug`,Icon:te},{id:`usage`,tkey:`nav.usage`,Icon:ne},{id:`codex-auth`,tkey:`nav.codexAuth`,Icon:L},{id:`api`,tkey:`nav.api`,Icon:me}],Zt={light:de,dark:fe,system:pe},Qt={light:`theme.light`,dark:`theme.dark`,system:`theme.system`};function $t(e){if(!e||typeof e!=`object`||!(`version`in e))return null;let t=e.version;return typeof t==`string`&&t.length>0?t:null}function en(){let e=localStorage.getItem(Yt);return e===`light`||e===`dark`?e:`system`}function tn(){let[e,t]=(0,_.useState)(qt),n=e=>{location.hash=e,t(e)},[r,i]=(0,_.useState)(en),[a,o]=(0,_.useState)(null),{locale:s,setLocale:c}=Ce(),l=we();(0,_.useEffect)(()=>{let e=()=>t(qt());return window.addEventListener(`hashchange`,e),()=>window.removeEventListener(`hashchange`,e)},[]),(0,_.useEffect)(()=>{let e=document.documentElement;r===`system`?(e.removeAttribute(`data-theme`),localStorage.removeItem(Yt)):(e.setAttribute(`data-theme`,r),localStorage.setItem(Yt,r))},[r]),(0,_.useEffect)(()=>{let e=!1,t=async()=>{try{let t=await fetch(`${Jt}/healthz`);if(!t.ok)return;let n=$t(await t.json());!e&&n&&o(n)}catch{}};t();let n=setInterval(t,3e4);return()=>{e=!0,clearInterval(n)}},[]);let u=()=>i(e=>e===`light`?`dark`:e===`dark`?`system`:`light`),d=Zt[r],f=a??`2.7.2`,[p,m]=(0,_.useState)(!1);return(0,S.jsxs)(`div`,{className:`app`,children:[(0,S.jsxs)(`aside`,{className:`sidebar`,children:[(0,S.jsxs)(`div`,{className:`brand`,children:[(0,S.jsx)(`span`,{className:`brand-logo`,role:`img`,"aria-label":`opencodex logo`}),(0,S.jsx)(`span`,{className:`name`,children:`opencodex`}),(0,S.jsxs)(`span`,{className:`ver`,children:[`v`,f]})]}),(0,S.jsx)(`nav`,{children:Xt.map(({id:t,tkey:r,Icon:i})=>(0,S.jsxs)(`button`,{className:`nav-item${e===t?` active`:``}`,"data-page":t,onClick:()=>n(t),"aria-current":e===t?`page`:void 0,children:[(0,S.jsx)(i,{}),` `,l(r)]},t))}),(0,S.jsxs)(`div`,{className:`sidebar-foot`,children:[(0,S.jsxs)(`div`,{className:`lang-toggle`,children:[(0,S.jsx)(me,{"aria-hidden":!0}),(0,S.jsx)(Ae,{value:s,options:_e.map(e=>({value:e.code,label:e.name})),onChange:e=>c(e),label:l(`lang.label`),placement:`right`,style:{flex:1,minWidth:0,width:`100%`}})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle`,onClick:u,"aria-label":`${l(`theme.label`)}: ${l(Qt[r])}`,title:`${l(`theme.label`)}: ${l(Qt[r])}`,children:[(0,S.jsx)(d,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(Qt[r])})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle stop-toggle`,onClick:async()=>{if(confirm(l(`dash.stopConfirm`))){m(!0);try{await fetch(`${Jt}/api/stop`,{method:`POST`})}catch{}}},disabled:p,"aria-label":l(`dash.stop`),title:l(`dash.stop`),children:[(0,S.jsx)(I,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(p?`dash.stopping`:`dash.stop`)})]}),(0,S.jsxs)(`a`,{className:`sidebar-link`,href:`https://github.com/lidge-jun/opencodex`,target:`_blank`,rel:`noreferrer`,children:[(0,S.jsx)(F,{}),` `,l(`common.github`)]})]})]}),(0,S.jsx)(`main`,{className:`main`,children:(0,S.jsxs)(`div`,{className:`main-inner`,children:[e===`dashboard`&&(0,S.jsx)(He,{apiBase:Jt}),e===`providers`&&(0,S.jsx)(nt,{apiBase:Jt}),e===`models`&&(0,S.jsx)(ct,{apiBase:Jt}),e===`subagents`&&(0,S.jsx)(lt,{apiBase:Jt}),e===`logs`&&(0,S.jsx)(yt,{apiBase:Jt}),e===`debug`&&(0,S.jsx)(bt,{apiBase:Jt}),e===`usage`&&(0,S.jsx)(Dt,{apiBase:Jt}),e===`codex-auth`&&(0,S.jsx)(kt,{apiBase:Jt}),e===`api`&&(0,S.jsx)(Pt,{apiBase:Jt})]})})]})}v.createRoot(document.getElementById(`root`)).render((0,S.jsx)(_.StrictMode,{children:(0,S.jsx)(Se,{children:(0,S.jsx)(tn,{})})}));
+  }'`})]})]})}var Ft=`opencodex-api-token`,It=!1,Lt=null;function Rt(e){try{let t=e instanceof Request?e.url:String(e);return new URL(t,window.location.href).pathname}catch{return null}}function zt(e){let t=Rt(e);return!!t&&(t.startsWith(`/api/`)||t.startsWith(`/v1/`))}function Bt(){try{return sessionStorage.getItem(Ft)?.trim()||null}catch{return null}}function Vt(e){try{sessionStorage.setItem(Ft,e)}catch{}}function Ht(){try{sessionStorage.removeItem(Ft)}catch{}}function Ut(e,t,n){let r=new Headers(t?.headers??(e instanceof Request?e.headers:void 0));return r.set(`X-OpenCodex-API-Key`,n),e instanceof Request?[new Request(e,{headers:r}),t?{...t,headers:r}:void 0]:[e,{...t,headers:r}]}async function Wt(){return Lt||(Lt=Promise.resolve().then(()=>window.prompt(`OpenCodex API token`)?.trim()||null).finally(()=>{Lt=null}),Lt)}function Gt(){if(It)return;It=!0;let e=window.fetch.bind(window);window.fetch=async(t,n)=>{if(!zt(t))return e(t,n);let r=Bt(),[i,a]=r?Ut(t,n,r):[t,n],o=await e(i,a);if(o.status!==401)return o;r&&Ht();let s=await Wt();if(!s)return o;Vt(s);let[c,l]=Ut(t,n,s),u=await e(c,l);return u.status===401&&Ht(),u}}Gt();var Kt=new Set([`dashboard`,`providers`,`models`,`subagents`,`logs`,`debug`,`usage`,`codex-auth`,`api`]);function qt(){let e=location.hash.replace(/^#\/?/,``);return Kt.has(e)?e:`dashboard`}var Jt=``,Yt=`ocx-theme`,Xt=[{id:`dashboard`,tkey:`nav.dashboard`,Icon:w},{id:`providers`,tkey:`nav.providers`,Icon:T},{id:`models`,tkey:`nav.models`,Icon:E},{id:`subagents`,tkey:`nav.subagents`,Icon:ee},{id:`logs`,tkey:`nav.logs`,Icon:D},{id:`debug`,tkey:`nav.debug`,Icon:te},{id:`usage`,tkey:`nav.usage`,Icon:ne},{id:`codex-auth`,tkey:`nav.codexAuth`,Icon:L},{id:`api`,tkey:`nav.api`,Icon:me}],Zt={light:de,dark:fe,system:pe},Qt={light:`theme.light`,dark:`theme.dark`,system:`theme.system`};function $t(e){if(!e||typeof e!=`object`||!(`version`in e))return null;let t=e.version;return typeof t==`string`&&t.length>0?t:null}function en(){let e=localStorage.getItem(Yt);return e===`light`||e===`dark`?e:`system`}function tn(){let[e,t]=(0,_.useState)(qt),n=e=>{location.hash=e,t(e)},[r,i]=(0,_.useState)(en),[a,o]=(0,_.useState)(null),{locale:s,setLocale:c}=Ce(),l=we();(0,_.useEffect)(()=>{let e=()=>t(qt());return window.addEventListener(`hashchange`,e),()=>window.removeEventListener(`hashchange`,e)},[]),(0,_.useEffect)(()=>{let e=document.documentElement;r===`system`?(e.removeAttribute(`data-theme`),localStorage.removeItem(Yt)):(e.setAttribute(`data-theme`,r),localStorage.setItem(Yt,r))},[r]),(0,_.useEffect)(()=>{let e=!1,t=async()=>{try{let t=await fetch(`${Jt}/healthz`);if(!t.ok)return;let n=$t(await t.json());!e&&n&&o(n)}catch{}};t();let n=setInterval(t,3e4);return()=>{e=!0,clearInterval(n)}},[]);let u=()=>i(e=>e===`light`?`dark`:e===`dark`?`system`:`light`),d=Zt[r],f=a??`2.7.3`,[p,m]=(0,_.useState)(!1);return(0,S.jsxs)(`div`,{className:`app`,children:[(0,S.jsxs)(`aside`,{className:`sidebar`,children:[(0,S.jsxs)(`div`,{className:`brand`,children:[(0,S.jsx)(`span`,{className:`brand-logo`,role:`img`,"aria-label":`opencodex logo`}),(0,S.jsx)(`span`,{className:`name`,children:`opencodex`}),(0,S.jsxs)(`span`,{className:`ver`,children:[`v`,f]})]}),(0,S.jsx)(`nav`,{children:Xt.map(({id:t,tkey:r,Icon:i})=>(0,S.jsxs)(`button`,{className:`nav-item${e===t?` active`:``}`,"data-page":t,onClick:()=>n(t),"aria-current":e===t?`page`:void 0,children:[(0,S.jsx)(i,{}),` `,l(r)]},t))}),(0,S.jsxs)(`div`,{className:`sidebar-foot`,children:[(0,S.jsxs)(`div`,{className:`lang-toggle`,children:[(0,S.jsx)(me,{"aria-hidden":!0}),(0,S.jsx)(Ae,{value:s,options:_e.map(e=>({value:e.code,label:e.name})),onChange:e=>c(e),label:l(`lang.label`),placement:`right`,style:{flex:1,minWidth:0,width:`100%`}})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle`,onClick:u,"aria-label":`${l(`theme.label`)}: ${l(Qt[r])}`,title:`${l(`theme.label`)}: ${l(Qt[r])}`,children:[(0,S.jsx)(d,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(Qt[r])})]}),(0,S.jsxs)(`button`,{type:`button`,className:`theme-toggle stop-toggle`,onClick:async()=>{if(confirm(l(`dash.stopConfirm`))){m(!0);try{await fetch(`${Jt}/api/stop`,{method:`POST`})}catch{}}},disabled:p,"aria-label":l(`dash.stop`),title:l(`dash.stop`),children:[(0,S.jsx)(I,{}),` `,(0,S.jsx)(`span`,{className:`mode`,children:l(p?`dash.stopping`:`dash.stop`)})]}),(0,S.jsxs)(`a`,{className:`sidebar-link`,href:`https://github.com/lidge-jun/opencodex`,target:`_blank`,rel:`noreferrer`,children:[(0,S.jsx)(F,{}),` `,l(`common.github`)]})]})]}),(0,S.jsx)(`main`,{className:`main`,children:(0,S.jsxs)(`div`,{className:`main-inner`,children:[e===`dashboard`&&(0,S.jsx)(He,{apiBase:Jt}),e===`providers`&&(0,S.jsx)(nt,{apiBase:Jt}),e===`models`&&(0,S.jsx)(ct,{apiBase:Jt}),e===`subagents`&&(0,S.jsx)(lt,{apiBase:Jt}),e===`logs`&&(0,S.jsx)(yt,{apiBase:Jt}),e===`debug`&&(0,S.jsx)(bt,{apiBase:Jt}),e===`usage`&&(0,S.jsx)(Dt,{apiBase:Jt}),e===`codex-auth`&&(0,S.jsx)(kt,{apiBase:Jt}),e===`api`&&(0,S.jsx)(Pt,{apiBase:Jt})]})})]})}v.createRoot(document.getElementById(`root`)).render((0,S.jsx)(_.StrictMode,{children:(0,S.jsx)(Se,{children:(0,S.jsx)(tn,{})})}));

package/gui/dist/index.html

@@ -16,7 +16,7 @@
         } catch (e) {}
       })();
     </script>
-    <script type="module" crossorigin src="/assets/index-DGFbiUtS.js"></script>
+    <script type="module" crossorigin src="/assets/index-DT4C-vKW.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-D7o1qwy-.css">
   </head>
   <body>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bitkyc08/opencodex",
-  "version": "2.7.2",
+  "version": "2.7.3",
   "description": "Universal provider proxy for OpenAI Codex — use any LLM with Codex CLI/App/SDK",
   "type": "module",
   "main": "./bin/package-main.mjs",

package/src/server/responses.ts

@@ -133,25 +133,12 @@ export function collabSurface(parsed: OcxParsedRequest): "v1" | "v2" | null {
 /**
  * Multi-agent guidance for this turn, or null when nothing applies.
  *
- * codex-rs only emits its Proactive delegation developer message on the v2 surface,
- * so when a v1-surface turn arrives at the synthetic top tier (codex converts
- * ultra -> max on the wire, so max arrival means the user picked the top rung) the
- * proxy supplies the same one-liner, wrapped in codex's own <multi_agent_mode> tags
- * (v1 turns never carry that fragment, so there is nothing to collide with).
- * Ultra is always advertised, so the guidance fires regardless of the multi_agent_v2
- * toggle.
- *
- * Dynamic model injection: when the user has configured a specific injectionModel,
- * the prompt names it so the agent knows which routed model to delegate to.
- *
- * Effort gate relaxation: when an injectionModel is set, the prompt fires at every
- * effort level, not just max/ultra — the user opted into delegation.
- *
- * Reasoning-effort injection: when an injectionEffort is configured alongside the
- * model, the prompt also tells the agent to pass `reasoning_effort` in spawn_agent
- * calls (codex-rs validates spawn efforts by catalog membership; unsupported rungs
- * are clamped on the wire). An effort WITHOUT a model changes nothing — the gate
- * and the base prompt stay exactly as before.
+ * V1 surface: codex-rs only emits its Proactive delegation developer message on the
+ * v2 surface, so when a v1-surface turn arrives at the synthetic top tier (codex
+ * converts ultra -> max on the wire, so max arrival means the user picked the top
+ * rung) the proxy supplies the same one-liner, wrapped in codex's own
+ * <multi_agent_mode> tags. That is ALL v1 gets — no model designation, no roster
+ * (kept lean by request, devlog 260710): ultra-tier injection is sufficient there.
  *
  * V2 surface (flat spawn_agent — sol/terra under the default mode, EVERY model when
  * `ocx v2 mode v2` forces the pins): codex-rs already emits its own Proactive text
@@ -169,16 +156,12 @@ export function collabSurface(parsed: OcxParsedRequest): "v1" | "v2" | null {
  * regardless of the flag (spawn.rs), so the prompt tells the model to pass the
  * arguments even though the schema does not list them.
  *
- * Prompt structure follows the OpenAI prompt-engineering guidance (developers.openai.com):
- * explicit rule lists first, a concrete few-shot example call, reference context (the
- * roster) last. A user-configured `injectionPrompt` replaces the built-in body with
- * {{model}}/{{effort}}/{{roster}} placeholder substitution; firing gates are unchanged.
- *
- * Featured roster: whenever guidance fires (either surface), the configured
- * `subagentModels` (the 5-list the catalog features for spawn_agent) are appended
- * with the effort ladder each advertises in the injected catalog — the same list
- * codex-rs validates spawn efforts against — so the agent can pick per-task models
- * and honest effort rungs instead of flying blind behind the hidden schema.
+ * The v2 body is budgeted to <= 700 chars (V2_GUIDANCE_CHAR_BUDGET): rules first,
+ * then the preferred model, then the compact roster of configured `subagentModels`
+ * with the effort ladder each advertises in the injected catalog (the list codex-rs
+ * validates spawn efforts against). A user-configured `injectionPrompt` replaces the
+ * v2 body with {{model}}/{{effort}}/{{roster}} placeholder substitution (own length,
+ * user-owned); firing gates are unchanged.
  */
 export async function multiAgentGuidanceText(parsed: OcxParsedRequest, injectionModel?: string, injectionEffort?: string, subagentModels?: string[], injectionPrompt?: string): Promise<string | null> {
   const surface = collabSurface(parsed);
@@ -193,60 +176,33 @@ export async function multiAgentGuidanceText(parsed: OcxParsedRequest, injection
     if (injectionPrompt) {
       return `<multi_agent_mode>${applyInjectionPlaceholders(injectionPrompt, injectionModel, injectionEffort, roster)}</multi_agent_mode>`;
     }
-    const rules = [
-      "- This spawn_agent tool ACCEPTS optional \"model\" and \"reasoning_effort\" string arguments even though its published schema does not list them. The runtime parses and applies both. Never claim sub-agent models cannot be selected.",
-      "- When the user names a sub-agent model, pass it in the \"model\" argument exactly as given.",
-      "- Any spawn_agent call that sets \"model\" or \"reasoning_effort\" MUST also set \"fork_turns\" to \"none\" (or a positive integer string such as \"3\" for a partial fork); overrides are rejected on a full-history fork.",
-      "- With fork_turns \"none\", the child receives NO parent context: write a fully self-contained task message with paths, goals, and constraints.",
-    ];
+    let text = "spawn_agent also accepts hidden \"model\" and \"reasoning_effort\" string arguments "
+      + "(not in the schema, but parsed and applied) — never claim sub-agent models cannot be selected. "
+      + "When setting either, set fork_turns to \"none\" (or e.g. \"3\"; full-history forks reject overrides) "
+      + "and make the message self-contained.";
     if (injectionModel) {
-      rules.push(`- A preferred sub-agent model is configured: "${injectionModel}". Use it for independent sub-tasks unless the user explicitly asks for another model.`);
+      text += ` Preferred sub-agent: model "${injectionModel}"`
+        + (injectionEffort ? `, reasoning_effort "${injectionEffort}"` : "")
+        + " — use it unless the user names another.";
     }
-    if (injectionEffort) {
-      rules.push(`- A preferred sub-agent reasoning effort is configured: "${injectionEffort}". Pass it in the "reasoning_effort" argument of those spawn_agent calls.`);
+    text += roster;
+    if (text.length > V2_GUIDANCE_CHAR_BUDGET) {
+      // Roster is the only unbounded part — drop it before breaking the budget.
+      text = text.slice(0, text.length - roster.length);
     }
-    const exampleModel = injectionModel ?? "gpt-5.6-terra";
-    const example = "## Example spawn_agent call\n"
-      + "```json\n"
-      + JSON.stringify({
-        task_name: "example_task",
-        message: "Self-contained task description with paths, goals, and constraints.",
-        fork_turns: "none",
-        model: exampleModel,
-        ...(injectionEffort ? { reasoning_effort: injectionEffort } : {}),
-      }, null, 2)
-      + "\n```";
-    const text = `## Sub-agent model selection rules\n${rules.join("\n")}\n\n${example}${roster}`;
     return `<multi_agent_mode>${text}</multi_agent_mode>`;
   }
 
   const effort = parsed.options.reasoning;
-  // When the user has selected a specific injection model, fire the delegation prompt
-  // at ANY effort level. Otherwise preserve the original gate: top tier only (max/ultra).
-  if (!injectionModel && effort !== "max" && effort !== "ultra") return null;
-
-  if (injectionPrompt) {
-    const roster = await subagentRosterText(subagentModels);
-    return `<multi_agent_mode>${applyInjectionPlaceholders(injectionPrompt, injectionModel, injectionEffort, roster)}</multi_agent_mode>`;
-  }
-
-  let text = PROACTIVE_MULTI_AGENT_MODE_TEXT;
-
-  // Append the selected model when the user has configured a specific injection target.
-  if (injectionModel) {
-    text += `\n\nA preferred sub-agent model is configured: "${injectionModel}". `
-      + `When delegating, call spawn_agent and set its model argument to exactly "${injectionModel}". `
-      + "Use it for independent sub-tasks unless the user explicitly asks for another model.";
-    if (injectionEffort) {
-      text += ` A preferred sub-agent reasoning effort is also configured: "${injectionEffort}". `
-        + `Set the reasoning_effort argument of spawn_agent to exactly "${injectionEffort}" for those sub-agents.`;
-    }
-  }
-
-  text += await subagentRosterText(subagentModels);
-  return `<multi_agent_mode>${text}</multi_agent_mode>`;
+  // v1 keeps only the upstream-parity behavior: Proactive text at the top tier
+  // (ultra arrives as max on the wire). No designation/roster payload here.
+  if (effort !== "max" && effort !== "ultra") return null;
+  return `<multi_agent_mode>${PROACTIVE_MULTI_AGENT_MODE_TEXT}</multi_agent_mode>`;
 }
 
+/** Hard budget for the built-in v2 guidance body (user request: keep injection lean). */
+export const V2_GUIDANCE_CHAR_BUDGET = 700;
+
 /** {{model}}/{{effort}}/{{roster}} substitution for the user-configured injectionPrompt. */
 function applyInjectionPlaceholders(prompt: string, model?: string, effort?: string, roster?: string): string {
   return prompt
@@ -256,7 +212,7 @@ function applyInjectionPlaceholders(prompt: string, model?: string, effort?: str
 }
 
 /**
- * "\n\nOther available sub-agent models..." roster block, or "" when no configured
+ * Compact one-line roster of configured sub-agent models, or "" when no configured
  * model resolves to a catalog entry. Efforts come from the injected catalog
  * (catalogModelEfforts) so only rungs codex-rs will actually accept are advertised.
  */
@@ -265,13 +221,17 @@ async function subagentRosterText(subagentModels?: string[]): Promise<string> {
   if (featured.length === 0) return "";
   const { catalogModelEfforts } = await import("../codex/catalog");
   const efforts = catalogModelEfforts(featured);
-  const lines = featured
-    .filter(id => efforts.has(id))
-    .map(id => `- "${id}" (reasoning_effort options: ${efforts.get(id)!.join(", ")})`);
-  if (lines.length === 0) return "";
-  return "\n\nConfigured sub-agent model roster (valid values for spawn_agent's \"model\" argument, "
-    + "with the reasoning_effort each supports):\n"
-    + lines.join("\n");
+  const resolved = featured.filter(id => efforts.has(id));
+  if (resolved.length === 0) return "";
+  const ladders = new Set(resolved.map(id => efforts.get(id)!.join("/")));
+  if (ladders.size === 1) {
+    // Shared ladder (the common case: the injected catalog advertises one rung set)
+    // -> state it once instead of per model, keeping the roster inside the budget.
+    const ids = resolved.map(id => `"${id}"`).join(", ");
+    return ` Available models (reasoning_effort ${[...ladders][0]}): ${ids}.`;
+  }
+  const entries = resolved.map(id => `"${id}" (${efforts.get(id)!.join("/")})`);
+  return ` Available models (valid reasoning_effort): ${entries.join(", ")}.`;
 }
 
 /**