@srouterhq/server 0.0.0-stage → 0.9.0

package/dist/lib/budget.js

@@ -0,0 +1,40 @@
+// Monthly free-tier budgets are stored as human labels like '~120M', '~50-100M',
+// '~12M', or '~500K'. Parse the upper bound to an absolute token count for
+// quota math (headroom guardrail, token-usage bar). Returns 0 for unknown/empty
+// labels, which callers treat as "no budget info".
+export function parseBudget(s) {
+    if (!s)
+        return 0;
+    // Require a magnitude unit (M/K). A bare number with no unit is a rate limit
+    // or placeholder, not a monthly token budget — "free · 40 RPM",
+    // "free · 200/hr per IP", "promo (trial)", "~? (anon)" — so treat those as
+    // "no budget info" (0), per this function's contract. Without the required
+    // unit the old regex parsed "free · 40 RPM" as 40 tokens, which showed a bogus
+    // budget and made the headroom guardrail penalize the model after one request.
+    const m = s.match(/~?([\d.]+)(?:-([\d.]+))?([MK])/);
+    if (!m)
+        return 0;
+    const high = parseFloat(m[2] ?? m[1]);
+    if (Number.isNaN(high))
+        return 0;
+    const unit = m[3] === 'M' ? 1_000_000 : 1_000;
+    return high * unit;
+}
+// Sort score for the "budget" preset on the fallback/profiles dashboards.
+// Shares parseBudget so rate-limit labels can't masquerade as budgets: the
+// per-route copies this replaced multiplied the bare number in
+// "free · 40 RPM" by 1e6 because the 'M' in "RPM" hit the unit check, ranking
+// a 40-requests-per-minute model as a 40-million-token budget.
+// tpd_limit still wins when present (a concrete daily cap beats a "~" label),
+// and "unlimited"/"∞" labels keep sorting above every parsed budget.
+export function monthlyBudgetScore(m) {
+    if (m.tpd_limit != null)
+        return m.tpd_limit * 30;
+    const str = m.monthly_token_budget;
+    if (!str)
+        return 0;
+    if (str.toLowerCase().includes('unlimited') || str.includes('∞'))
+        return Infinity;
+    return parseBudget(str);
+}
+//# sourceMappingURL=budget.js.map

package/dist/lib/budget.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"budget.js","sourceRoot":"","sources":["../../src/lib/budget.ts"],"names":[],"mappings":"AAAA,iFAAiF;AACjF,2EAA2E;AAC3E,gFAAgF;AAChF,mDAAmD;AACnD,MAAM,UAAU,WAAW,CAAC,CAAS;IACnC,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,6EAA6E;IAC7E,gEAAgE;IAChE,2EAA2E;IAC3E,2EAA2E;IAC3E,+EAA+E;IAC/E,+EAA+E;IAC/E,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,IAAI,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACjB,MAAM,IAAI,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACtC,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IACjC,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC;IAC9C,OAAO,IAAI,GAAG,IAAI,CAAC;AACrB,CAAC;AAED,0EAA0E;AAC1E,2EAA2E;AAC3E,+DAA+D;AAC/D,8EAA8E;AAC9E,+DAA+D;AAC/D,8EAA8E;AAC9E,qEAAqE;AACrE,MAAM,UAAU,kBAAkB,CAAC,CAA6D;IAC9F,IAAI,CAAC,CAAC,SAAS,IAAI,IAAI;QAAE,OAAO,CAAC,CAAC,SAAS,GAAG,EAAE,CAAC;IACjD,MAAM,GAAG,GAAG,CAAC,CAAC,oBAAoB,CAAC;IACnC,IAAI,CAAC,GAAG;QAAE,OAAO,CAAC,CAAC;IACnB,IAAI,GAAG,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC;QAAE,OAAO,QAAQ,CAAC;IAClF,OAAO,WAAW,CAAC,GAAG,CAAC,CAAC;AAC1B,CAAC"}

package/dist/lib/client-classifier.d.ts

@@ -0,0 +1,10 @@
+import type { Request } from 'express';
+export declare const CLIENT_AGENTS: readonly ["claude-code", "codex", "cline", "roo", "continue", "aider", "opencode", "goose", "qwen-code", "kilo-code", "crush", "deepseek-harness", "mimo-code", "atomcode", "openclaw", "hermes-agent", "pi", "cursor", "gemini-cli", "zed", "jetbrains", "ollama-client", "openai-sdk", "anthropic-sdk", "unknown"];
+export type ClientAgent = (typeof CLIENT_AGENTS)[number];
+/**
+ * Best-effort classifier from stable protocol/header signals first, then UA.
+ * It intentionally returns a coarse enum: analytics should remain useful when
+ * a client revs its version string or wraps an underlying SDK.
+ */
+export declare function classifyClientAgent(req: Request): ClientAgent;
+//# sourceMappingURL=client-classifier.d.ts.map

package/dist/lib/client-classifier.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"client-classifier.d.ts","sourceRoot":"","sources":["../../src/lib/client-classifier.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAEvC,eAAO,MAAM,aAAa,sTA0BhB,CAAC;AAEX,MAAM,MAAM,WAAW,GAAG,CAAC,OAAO,aAAa,CAAC,CAAC,MAAM,CAAC,CAAC;AAOzD;;;;GAIG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,OAAO,GAAG,WAAW,CA+D7D"}

package/dist/lib/client-classifier.js

@@ -0,0 +1,126 @@
+export const CLIENT_AGENTS = [
+    'claude-code',
+    'codex',
+    'cline',
+    'roo',
+    'continue',
+    'aider',
+    'opencode',
+    'goose',
+    'qwen-code',
+    'kilo-code',
+    'crush',
+    'deepseek-harness',
+    'mimo-code',
+    'atomcode',
+    'openclaw',
+    'hermes-agent',
+    'pi',
+    'cursor',
+    'gemini-cli',
+    'zed',
+    'jetbrains',
+    'ollama-client',
+    'openai-sdk',
+    'anthropic-sdk',
+    'unknown',
+];
+function header(req, name) {
+    const raw = req.headers[name];
+    return (Array.isArray(raw) ? raw[0] : raw)?.toLowerCase() ?? '';
+}
+/**
+ * Best-effort classifier from stable protocol/header signals first, then UA.
+ * It intentionally returns a coarse enum: analytics should remain useful when
+ * a client revs its version string or wraps an underlying SDK.
+ */
+export function classifyClientAgent(req) {
+    const path = (req.originalUrl ?? req.url ?? '').split('?')[0].toLowerCase();
+    const tokenizedPath = path.replace(/^\/v1\/t\/[^/]+/, '');
+    const ua = header(req, 'user-agent');
+    // Strongest signal first: the dedicated session header only Claude Code sets.
+    if (header(req, 'x-claude-code-session-id'))
+        return 'claude-code';
+    // UA signals must beat wire-format fallbacks: Qwen Code speaks the Gemini
+    // wire on /v1beta, and a plain openai SDK can call /v1/responses — neither
+    // should be counted as the surface's flagship client. Claude Code's real UA
+    // is `claude-cli/x.y.z (external, cli)`.
+    if (/claude[- ]?code|\bclaude-cli\b/.test(ua))
+        return 'claude-code';
+    if (/\bcodex\b/.test(ua))
+        return 'codex';
+    if (/qwen[- ]?code|\bqwen-cli\b/.test(ua))
+        return 'qwen-code';
+    if (/gemini[- /]?cli/.test(ua))
+        return 'gemini-cli';
+    if (/kilo[- ]?code|\bkilocode\b/.test(ua))
+        return 'kilo-code';
+    if (/\bcrush(?:\/|\s|$)/.test(ua))
+        return 'crush';
+    // `deepseek-harness/<version> (+https://github.com/deepseek-ai/deepseek-harness)`
+    if (/deepseek[- ]?harness|\bdsh\//.test(ua))
+        return 'deepseek-harness';
+    // MiMo Code is an OpenCode derivative and can carry either name in its UA,
+    // so it has to be tested before the plain `opencode` rule below.
+    if (/mimo[- ]?code|\bmimo\//.test(ua))
+        return 'mimo-code';
+    // `atomcode/<version>`
+    if (/\batomcode\b/.test(ua))
+        return 'atomcode';
+    // OpenClaw names itself only through the static provider header our
+    // generator writes (`openclaw`); its own `openclaw/<version>` UA is reserved
+    // for endpoints it recognises.
+    if (/\bopenclaw\b/.test(ua))
+        return 'openclaw';
+    // Hermes Agent likewise only names itself to hosts it knows; the generator's
+    // `default_headers` sends `hermes-agent`, and its /v1/models probe sends
+    // `hermes-cli/<version>` on its own.
+    if (/hermes[- ]?agent|\bhermes-cli\b/.test(ua))
+        return 'hermes-agent';
+    // Pi likewise only names itself to hosts it knows; the generator's provider
+    // header sends `pi-coding-agent`, and Pi's own UA is
+    // `pi/<version> (<platform>; node/<version>; <arch>)`.
+    if (/\bpi-coding-agent\b|^pi\/\d/.test(ua))
+        return 'pi';
+    if (/opencode/.test(ua))
+        return 'opencode';
+    if (/\bcline\b/.test(ua))
+        return 'cline';
+    // The Roo Code extension and CLI send `RooCode/<version>`.
+    if (/\broo[- /]/.test(ua) || /roo ?code/.test(ua))
+        return 'roo';
+    if (/continue(?:\.dev)?/.test(ua))
+        return 'continue';
+    if (/\baider\b/.test(ua))
+        return 'aider';
+    if (/\bgoose\b/.test(ua))
+        return 'goose';
+    if (/\bcursor\b/.test(ua))
+        return 'cursor';
+    if (/\bzed\b/.test(ua))
+        return 'zed';
+    if (/jetbrains|intellij|pycharm|webstorm/.test(ua))
+        return 'jetbrains';
+    if (/\bollama\b/.test(ua))
+        return 'ollama-client';
+    if (/anthropic|claude-sdk/.test(ua))
+        return 'anthropic-sdk';
+    if (/openai|langchain|litellm/.test(ua))
+        return 'openai-sdk';
+    // Wire-format fallbacks for clients with no recognizable UA.
+    if (path.startsWith('/v1/responses') || tokenizedPath.startsWith('/responses'))
+        return 'codex';
+    if (path.startsWith('/v1beta/'))
+        return 'gemini-cli';
+    if (path.startsWith('/v1/messages')) {
+        // anthropic.ts session affinity treats x-session-id as a Claude Code
+        // signal on this surface; anything else Anthropic-shaped stays coarse.
+        return header(req, 'x-session-id') ? 'claude-code' : 'anthropic-sdk';
+    }
+    if (tokenizedPath.startsWith('/api/chat')
+        || tokenizedPath.startsWith('/api/generate')
+        || tokenizedPath.startsWith('/api/tags'))
+        return 'ollama-client';
+    return 'unknown';
+}
+//# sourceMappingURL=client-classifier.js.map

package/dist/lib/client-classifier.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"client-classifier.js","sourceRoot":"","sources":["../../src/lib/client-classifier.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,aAAa,GAAG;IAC3B,aAAa;IACb,OAAO;IACP,OAAO;IACP,KAAK;IACL,UAAU;IACV,OAAO;IACP,UAAU;IACV,OAAO;IACP,WAAW;IACX,WAAW;IACX,OAAO;IACP,kBAAkB;IAClB,WAAW;IACX,UAAU;IACV,UAAU;IACV,cAAc;IACd,IAAI;IACJ,QAAQ;IACR,YAAY;IACZ,KAAK;IACL,WAAW;IACX,eAAe;IACf,YAAY;IACZ,eAAe;IACf,SAAS;CACD,CAAC;AAIX,SAAS,MAAM,CAAC,GAAY,EAAE,IAAY;IACxC,MAAM,GAAG,GAAG,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC9B,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,WAAW,EAAE,IAAI,EAAE,CAAC;AAClE,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAY;IAC9C,MAAM,IAAI,GAAG,CAAC,GAAG,CAAC,WAAW,IAAI,GAAG,CAAC,GAAG,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;IAC5E,MAAM,aAAa,GAAG,IAAI,CAAC,OAAO,CAAC,iBAAiB,EAAE,EAAE,CAAC,CAAC;IAC1D,MAAM,EAAE,GAAG,MAAM,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;IAErC,8EAA8E;IAC9E,IAAI,MAAM,CAAC,GAAG,EAAE,0BAA0B,CAAC;QAAE,OAAO,aAAa,CAAC;IAElE,0EAA0E;IAC1E,2EAA2E;IAC3E,4EAA4E;IAC5E,yCAAyC;IACzC,IAAI,gCAAgC,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,aAAa,CAAC;IACpE,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,OAAO,CAAC;IACzC,IAAI,4BAA4B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,WAAW,CAAC;IAC9D,IAAI,iBAAiB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,YAAY,CAAC;IACpD,IAAI,4BAA4B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,WAAW,CAAC;IAC9D,IAAI,oBAAoB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,OAAO,CAAC;IAClD,kFAAkF;IAClF,IAAI,8BAA8B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,kBAAkB,CAAC;IACvE,2EAA2E;IAC3E,iEAAiE;IACjE,IAAI,wBAAwB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,WAAW,CAAC;IAC1D,uBAAuB;IACvB,IAAI,cAAc,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,UAAU,CAAC;IAC/C,oEAAoE;IACpE,6EAA6E;IAC7E,+BAA+B;IAC/B,IAAI,cAAc,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,UAAU,CAAC;IAC/C,6EAA6E;IAC7E,yEAAyE;IACzE,qCAAqC;IACrC,IAAI,iCAAiC,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,cAAc,CAAC;IACtE,4EAA4E;IAC5E,qDAAqD;IACrD,uDAAuD;IACvD,IAAI,6BAA6B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IACxD,IAAI,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,UAAU,CAAC;IAC3C,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,OAAO,CAAC;IACzC,2DAA2D;IAC3D,IAAI,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,KAAK,CAAC;IAChE,IAAI,oBAAoB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,UAAU,CAAC;IACrD,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,OAAO,CAAC;IACzC,IAAI,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,OAAO,CAAC;IACzC,IAAI,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,QAAQ,CAAC;IAC3C,IAAI,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,KAAK,CAAC;IACrC,IAAI,qCAAqC,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,WAAW,CAAC;IACvE,IAAI,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,eAAe,CAAC;IAClD,IAAI,sBAAsB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,eAAe,CAAC;IAC5D,IAAI,0BAA0B,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,YAAY,CAAC;IAE7D,6DAA6D;IAC7D,IAAI,IAAI,CAAC,UAAU,CAAC,eAAe,CAAC,IAAI,aAAa,CAAC,UAAU,CAAC,YAAY,CAAC;QAAE,OAAO,OAAO,CAAC;IAC/F,IAAI,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC;QAAE,OAAO,YAAY,CAAC;IACrD,IAAI,IAAI,CAAC,UAAU,CAAC,cAAc,CAAC,EAAE,CAAC;QACpC,qEAAqE;QACrE,uEAAuE;QACvE,OAAO,MAAM,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,eAAe,CAAC;IACvE,CAAC;IACD,IAAI,aAAa,CAAC,UAAU,CAAC,WAAW,CAAC;WAClC,aAAa,CAAC,UAAU,CAAC,eAAe,CAAC;WACzC,aAAa,CAAC,UAAU,CAAC,WAAW,CAAC;QAAE,OAAO,eAAe,CAAC;IACrE,OAAO,SAAS,CAAC;AACnB,CAAC"}

package/dist/lib/client-context.d.ts

@@ -0,0 +1,10 @@
+import type { NextFunction, Request, Response } from 'express';
+import { type ClientAgent } from './client-classifier.js';
+export interface ClientContext {
+    ip: string | null;
+    userAgent: string | null;
+    agent: ClientAgent | null;
+}
+export declare function clientContextMiddleware(req: Request, _res: Response, next: NextFunction): void;
+export declare function getClientContext(): ClientContext;
+//# sourceMappingURL=client-context.d.ts.map

package/dist/lib/client-context.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"client-context.d.ts","sourceRoot":"","sources":["../../src/lib/client-context.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,YAAY,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC;AAC/D,OAAO,EAAuB,KAAK,WAAW,EAAE,MAAM,wBAAwB,CAAC;AAE/E,MAAM,WAAW,aAAa;IAC5B,EAAE,EAAE,MAAM,GAAG,IAAI,CAAC;IAClB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,EAAE,WAAW,GAAG,IAAI,CAAC;CAC3B;AA0BD,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,YAAY,GAAG,IAAI,CAW9F;AAED,wBAAgB,gBAAgB,IAAI,aAAa,CAEhD"}

package/dist/lib/client-context.js

@@ -0,0 +1,39 @@
+import { AsyncLocalStorage } from 'async_hooks';
+import { classifyClientAgent } from './client-classifier.js';
+// Request-scoped caller identity, readable from anywhere below the middleware
+// without threading parameters through every logRequest() call site (the chat
+// proxy, responses, anthropic, fusion, embeddings and media paths all log).
+const storage = new AsyncLocalStorage();
+// Resolve the client IP. With Express's `trust proxy` disabled (the default),
+// this is the socket peer address and a spoofed X-Forwarded-For from a LAN
+// client is ignored. When TRUST_PROXY (#1024) opts into trusting a reverse
+// proxy, `req.ip` walks the configured trusted-proxy chain and returns the
+// first untrusted address — instead of trusting the leftmost caller-supplied
+// X-Forwarded-For value, which would let any direct caller spoof it.
+function resolveClientIp(req) {
+    const raw = req.ip ?? req.socket.remoteAddress ?? null;
+    // Normalize IPv4-mapped IPv6 ("::ffff:192.168.0.5" -> "192.168.0.5").
+    return raw?.replace(/^::ffff:/i, '') ?? null;
+}
+// Privacy opt-out: REQUEST_ANALYTICS_LOG_CLIENT=false stores nulls instead of
+// the caller's IP/UA. Read per request (not at module load) so tests and
+// embedders can toggle it without re-importing.
+function clientLoggingEnabled() {
+    return process.env.REQUEST_ANALYTICS_LOG_CLIENT !== 'false';
+}
+export function clientContextMiddleware(req, _res, next) {
+    if (!clientLoggingEnabled()) {
+        storage.run({ ip: null, userAgent: null, agent: null }, next);
+        return;
+    }
+    const ua = req.headers['user-agent'];
+    storage.run({
+        ip: resolveClientIp(req),
+        userAgent: typeof ua === 'string' ? ua.slice(0, 256) : null,
+        agent: classifyClientAgent(req),
+    }, next);
+}
+export function getClientContext() {
+    return storage.getStore() ?? { ip: null, userAgent: null, agent: null };
+}
+//# sourceMappingURL=client-context.js.map

package/dist/lib/client-context.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"client-context.js","sourceRoot":"","sources":["../../src/lib/client-context.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,aAAa,CAAC;AAEhD,OAAO,EAAE,mBAAmB,EAAoB,MAAM,wBAAwB,CAAC;AAQ/E,8EAA8E;AAC9E,8EAA8E;AAC9E,4EAA4E;AAC5E,MAAM,OAAO,GAAG,IAAI,iBAAiB,EAAiB,CAAC;AAEvD,8EAA8E;AAC9E,2EAA2E;AAC3E,2EAA2E;AAC3E,2EAA2E;AAC3E,6EAA6E;AAC7E,qEAAqE;AACrE,SAAS,eAAe,CAAC,GAAY;IACnC,MAAM,GAAG,GAAG,GAAG,CAAC,EAAE,IAAI,GAAG,CAAC,MAAM,CAAC,aAAa,IAAI,IAAI,CAAC;IACvD,sEAAsE;IACtE,OAAO,GAAG,EAAE,OAAO,CAAC,WAAW,EAAE,EAAE,CAAC,IAAI,IAAI,CAAC;AAC/C,CAAC;AAED,8EAA8E;AAC9E,yEAAyE;AACzE,gDAAgD;AAChD,SAAS,oBAAoB;IAC3B,OAAO,OAAO,CAAC,GAAG,CAAC,4BAA4B,KAAK,OAAO,CAAC;AAC9D,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,GAAY,EAAE,IAAc,EAAE,IAAkB;IACtF,IAAI,CAAC,oBAAoB,EAAE,EAAE,CAAC;QAC5B,OAAO,CAAC,GAAG,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE,IAAI,CAAC,CAAC;QAC9D,OAAO;IACT,CAAC;IACD,MAAM,EAAE,GAAG,GAAG,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;IACrC,OAAO,CAAC,GAAG,CAAC;QACV,EAAE,EAAE,eAAe,CAAC,GAAG,CAAC;QACxB,SAAS,EAAE,OAAO,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI;QAC3D,KAAK,EAAE,mBAAmB,CAAC,GAAG,CAAC;KAChC,EAAE,IAAI,CAAC,CAAC;AACX,CAAC;AAED,MAAM,UAAU,gBAAgB;IAC9B,OAAO,OAAO,CAAC,QAAQ,EAAE,IAAI,EAAE,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;AAC1E,CAAC"}

package/dist/lib/config.d.ts

@@ -0,0 +1,33 @@
+export interface Config {
+    port: number | string;
+    host: string;
+    dbPath: string | null;
+    dashboardOrigins: string[];
+    clientDist: string | null;
+    proxyRateLimitRpm: number;
+    /** JSON body limit (bytes) for the LLM wire surfaces — see
+     *  parseRequestBodyLimitBytes. REQUEST_BODY_LIMIT_MB overrides. */
+    requestBodyLimitBytes: number;
+    nodeEnv: string;
+    serveStaticAssets: boolean;
+    /**
+     * Express `trust proxy` setting, parsed from TRUST_PROXY (#1024).
+     * - false:        do not trust forwarded headers (default; direct callers
+     *                 cannot spoof X-Forwarded-For).
+     * - true:         trust all proxies (with a security warning).
+     * - number:       trust that many hops counted from the socket peer
+     *                 (Express's `trust proxy: 1` idiom for one proxy in front).
+     * - string[]:     trust only these proxy addresses/CIDRs.
+     */
+    trustProxy: boolean | number | string[];
+    /**
+     * Tri-state override for the CSP `upgrade-insecure-requests` directive (#682).
+     * - undefined: auto — emit the directive only when the request arrived over
+     *   TLS (or behind an HTTPS reverse proxy that forwarded X-Forwarded-Proto).
+     * - true:      always emit (force HTTPS upgrade even on plain HTTP).
+     * - false:     never emit (let HTTP LAN installs render the dashboard).
+     */
+    cspUpgradeInsecureRequests: boolean | undefined;
+}
+export declare function loadConfig(): Config;
+//# sourceMappingURL=config.d.ts.map

package/dist/lib/config.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"config.d.ts","sourceRoot":"","sources":["../../src/lib/config.ts"],"names":[],"mappings":"AAqCA,MAAM,WAAW,MAAM;IACrB,IAAI,EAAE,MAAM,GAAG,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,iBAAiB,EAAE,MAAM,CAAC;IAC1B;uEACmE;IACnE,qBAAqB,EAAE,MAAM,CAAC;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,iBAAiB,EAAE,OAAO,CAAC;IAC3B;;;;;;;;OAQG;IACH,UAAU,EAAE,OAAO,GAAG,MAAM,GAAG,MAAM,EAAE,CAAC;IACxC;;;;;;OAMG;IACH,0BAA0B,EAAE,OAAO,GAAG,SAAS,CAAC;CACjD;AAED,wBAAgB,UAAU,IAAI,MAAM,CAuBnC"}

package/dist/lib/config.js

@@ -0,0 +1,94 @@
+const DEFAULT_RPM = 120;
+// JSON body ceiling for the LLM wire surfaces (/v1, /v1beta, /mcp, Ollama
+// /api/*). Vision requests inline base64 images (~33% inflation over the raw
+// bytes; google.ts alone forwards images up to 8MB, ~10.7MB in base64), so a
+// single-screenshot Codex turn easily clears 10MB, and history replay stacks
+// several of them (an 11.85MB real-world request was observed). express.json
+// buffers the whole body in memory (2-3x transiently during the parse), so
+// this also bounds the per-request spike on small containers. Anything past
+// the ceiling 413s before routing — no fallback, no analytics row — so it
+// must sit above the largest payload we expect to serve. 25MB is the balance
+// point: it clears the observed 11.85MB replayed session with room to spare,
+// while keeping the worst case survivable on the small hosts this runs on (a
+// Pi or a 512MB VPS — the limit is PER REQUEST, so a handful of concurrent
+// maximal bodies is the number that actually has to fit in RAM). Inbound
+// image normalization shrinks payloads ~6-10x right after the parse, so the
+// steady state sits far below this. Raise REQUEST_BODY_LIMIT_MB on a bigger
+// box if a client genuinely sends more. The dashboard/admin surface keeps its
+// own smaller fixed ceiling.
+const DEFAULT_REQUEST_BODY_LIMIT_MB = 25;
+function parseRateLimitRpm() {
+    const raw = process.env.PROXY_RATE_LIMIT_RPM;
+    if (raw === undefined || raw.trim() === '')
+        return DEFAULT_RPM;
+    const n = Number(raw);
+    if (!Number.isFinite(n) || n < 0)
+        return DEFAULT_RPM;
+    return Math.floor(n);
+}
+function parseRequestBodyLimitBytes() {
+    const raw = process.env.REQUEST_BODY_LIMIT_MB;
+    if (raw === undefined || raw.trim() === '')
+        return DEFAULT_REQUEST_BODY_LIMIT_MB * 1024 * 1024;
+    const n = Number(raw);
+    if (!Number.isFinite(n) || n < 1)
+        return DEFAULT_REQUEST_BODY_LIMIT_MB * 1024 * 1024;
+    return Math.floor(n) * 1024 * 1024;
+}
+export function loadConfig() {
+    return {
+        port: process.env.PORT ?? 20268,
+        // Dual-stack ('::') by default so the dashboard is reachable over both IPv4
+        // and IPv6 (e.g. IPv6-enabled Docker networks — #180). Hosts with IPv6
+        // disabled fall back to IPv4-only below; HOST overrides the default outright.
+        host: process.env.HOST ?? '::',
+        dbPath: process.env.SROUTER_DB_PATH?.trim() || null,
+        dashboardOrigins: (process.env.DASHBOARD_ORIGINS ?? '')
+            .split(',')
+            .map(s => s.trim())
+            .filter(Boolean),
+        clientDist: process.env.CLIENT_DIST ?? null,
+        proxyRateLimitRpm: parseRateLimitRpm(),
+        requestBodyLimitBytes: parseRequestBodyLimitBytes(),
+        nodeEnv: process.env.NODE_ENV ?? 'development',
+        serveStaticAssets: true,
+        trustProxy: parseTrustProxy(),
+        // CSP_UPGRADE_INSECURE_REQUESTS: true|false forces the directive on/off;
+        // unset (the default) leaves it on auto — emit only when the request is
+        // already TLS or forwarded as https, so HTTP LAN installs still render.
+        cspUpgradeInsecureRequests: parseCspUpgradeInsecureRequests(),
+    };
+}
+// Parse TRUST_PROXY (#1024). unset/empty/false → false (do not trust forwarded
+// headers); true → trust all proxies; a non-negative integer → that many hops
+// from the socket peer (so `TRUST_PROXY=1` means "one proxy in front", the
+// Express idiom, not "trust everything"); otherwise a comma-separated list of
+// proxy addresses/CIDRs → trust only those hops.
+function parseTrustProxy() {
+    const raw = process.env.TRUST_PROXY;
+    if (raw === undefined || raw.trim() === '')
+        return false;
+    const lower = raw.trim().toLowerCase();
+    if (lower === 'true' || lower === 'yes')
+        return true;
+    if (lower === 'false' || lower === 'no')
+        return false;
+    if (/^\d+$/.test(lower)) {
+        const hops = Number(lower);
+        return hops > 0 ? hops : false;
+    }
+    const addrs = raw.split(',').map(s => s.trim()).filter(Boolean);
+    return addrs.length > 0 ? addrs : false;
+}
+function parseCspUpgradeInsecureRequests() {
+    const raw = process.env.CSP_UPGRADE_INSECURE_REQUESTS;
+    if (raw === undefined || raw.trim() === '')
+        return undefined;
+    const lower = raw.trim().toLowerCase();
+    if (lower === 'true' || lower === '1' || lower === 'yes')
+        return true;
+    if (lower === 'false' || lower === '0' || lower === 'no')
+        return false;
+    return undefined;
+}
+//# sourceMappingURL=config.js.map

package/dist/lib/config.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"config.js","sourceRoot":"","sources":["../../src/lib/config.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,GAAG,GAAG,CAAC;AAExB,0EAA0E;AAC1E,6EAA6E;AAC7E,6EAA6E;AAC7E,6EAA6E;AAC7E,6EAA6E;AAC7E,2EAA2E;AAC3E,4EAA4E;AAC5E,0EAA0E;AAC1E,6EAA6E;AAC7E,6EAA6E;AAC7E,6EAA6E;AAC7E,2EAA2E;AAC3E,yEAAyE;AACzE,4EAA4E;AAC5E,4EAA4E;AAC5E,8EAA8E;AAC9E,6BAA6B;AAC7B,MAAM,6BAA6B,GAAG,EAAE,CAAC;AAEzC,SAAS,iBAAiB;IACxB,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC;IAC7C,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE;QAAE,OAAO,WAAW,CAAC;IAC/D,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IACtB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,WAAW,CAAC;IACrD,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AACvB,CAAC;AAED,SAAS,0BAA0B;IACjC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC;IAC9C,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE;QAAE,OAAO,6BAA6B,GAAG,IAAI,GAAG,IAAI,CAAC;IAC/F,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IACtB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,6BAA6B,GAAG,IAAI,GAAG,IAAI,CAAC;IACrF,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,IAAI,GAAG,IAAI,CAAC;AACrC,CAAC;AAkCD,MAAM,UAAU,UAAU;IACxB,OAAO;QACL,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,KAAK;QAC/B,4EAA4E;QAC5E,uEAAuE;QACvE,8EAA8E;QAC9E,IAAI,EAAE,OAAO,CAAC,GAAG,CAAC,IAAI,IAAI,IAAI;QAC9B,MAAM,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,IAAI,EAAE,IAAI,IAAI;QACnD,gBAAgB,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,EAAE,CAAC;aACpD,KAAK,CAAC,GAAG,CAAC;aACV,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;aAClB,MAAM,CAAC,OAAO,CAAC;QAClB,UAAU,EAAE,OAAO,CAAC,GAAG,CAAC,WAAW,IAAI,IAAI;QAC3C,iBAAiB,EAAE,iBAAiB,EAAE;QACtC,qBAAqB,EAAE,0BAA0B,EAAE;QACnD,OAAO,EAAE,OAAO,CAAC,GAAG,CAAC,QAAQ,IAAI,aAAa;QAC9C,iBAAiB,EAAE,IAAI;QACvB,UAAU,EAAE,eAAe,EAAE;QAC7B,yEAAyE;QACzE,wEAAwE;QACxE,wEAAwE;QACxE,0BAA0B,EAAE,+BAA+B,EAAE;KAC9D,CAAC;AACJ,CAAC;AAED,+EAA+E;AAC/E,8EAA8E;AAC9E,2EAA2E;AAC3E,8EAA8E;AAC9E,iDAAiD;AACjD,SAAS,eAAe;IACtB,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC;IACpC,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE;QAAE,OAAO,KAAK,CAAC;IACzD,MAAM,KAAK,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvC,IAAI,KAAK,KAAK,MAAM,IAAI,KAAK,KAAK,KAAK;QAAE,OAAO,IAAI,CAAC;IACrD,IAAI,KAAK,KAAK,OAAO,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IACtD,IAAI,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QACxB,MAAM,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;QAC3B,OAAO,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC;IACjC,CAAC;IACD,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAChE,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC;AAC1C,CAAC;AAED,SAAS,+BAA+B;IACtC,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,6BAA6B,CAAC;IACtD,IAAI,GAAG,KAAK,SAAS,IAAI,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE;QAAE,OAAO,SAAS,CAAC;IAC7D,MAAM,KAAK,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACvC,IAAI,KAAK,KAAK,MAAM,IAAI,KAAK,KAAK,GAAG,IAAI,KAAK,KAAK,KAAK;QAAE,OAAO,IAAI,CAAC;IACtE,IAAI,KAAK,KAAK,OAAO,IAAI,KAAK,KAAK,GAAG,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IACvE,OAAO,SAAS,CAAC;AACnB,CAAC"}

package/dist/lib/content.d.ts

@@ -0,0 +1,33 @@
+import type { ChatMessage } from '@srouter/shared/types.js';
+export type ContentTextBlock = {
+    type: 'text';
+    text: string;
+};
+export type ContentBlock = ContentTextBlock | {
+    type: string;
+    [key: string]: unknown;
+};
+export declare function contentToString(content: unknown): string;
+export declare function estimateInputTokens(messages: ChatMessage[], tools?: readonly unknown[]): number;
+export declare function flattenMessageContent(messages: ChatMessage[]): ChatMessage[];
+export declare function contentHasImage(content: unknown): boolean;
+export declare function messageHasImage(messages: ChatMessage[]): boolean;
+export declare function stripImagesFromMessages(messages: ChatMessage[]): ChatMessage[];
+export declare const GITHUB_MAX_INPUT_TOKENS = 7500;
+/**
+ * Fit a message list inside GitHub Models' input ceiling before dispatch.
+ * Keeps the leading system prompt verbatim (its instructions are the whole
+ * point of prepending one) and then as many of the NEWEST messages as the
+ * remaining budget allows, dropping the oldest turns first — the same trade a
+ * client's own context window makes. A single message that blows the budget on
+ * its own is truncated rather than dropped, so the turn is never dispatched
+ * with nothing to answer.
+ *
+ * Returns the ORIGINAL array when the request already fits, which is both the
+ * common case and the cheap one: callers can hand every github-routed request
+ * through here without copying anything.
+ */
+export declare function truncateMessagesForGithub(messages: ChatMessage[], budget?: number): ChatMessage[];
+export declare function sanitizeResponse<T>(payload: T): T;
+export declare function normalizeOutboundContent<T>(payload: T): T;
+//# sourceMappingURL=content.d.ts.map

package/dist/lib/content.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"content.d.ts","sourceRoot":"","sources":["../../src/lib/content.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,0BAA0B,CAAC;AAU5D,MAAM,MAAM,gBAAgB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAC9D,MAAM,MAAM,YAAY,GAAG,gBAAgB,GAAG;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;CAAE,CAAC;AAEvF,wBAAgB,eAAe,CAAC,OAAO,EAAE,OAAO,GAAG,MAAM,CAoBxD;AASD,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,KAAK,CAAC,EAAE,SAAS,OAAO,EAAE,GAAG,MAAM,CAQ/F;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,WAAW,EAAE,GAAG,WAAW,EAAE,CAK5E;AAKD,wBAAgB,eAAe,CAAC,OAAO,EAAE,OAAO,GAAG,OAAO,CAMzD;AAID,wBAAgB,eAAe,CAAC,QAAQ,EAAE,WAAW,EAAE,GAAG,OAAO,CAEhE;AAID,wBAAgB,uBAAuB,CAAC,QAAQ,EAAE,WAAW,EAAE,GAAG,WAAW,EAAE,CAU9E;AAOD,eAAO,MAAM,uBAAuB,OAAO,CAAC;AAqB5C;;;;;;;;;;;;GAYG;AACH,wBAAgB,yBAAyB,CACvC,QAAQ,EAAE,WAAW,EAAE,EACvB,MAAM,GAAE,MAAgC,GACvC,WAAW,EAAE,CAyBf;AAcD,wBAAgB,gBAAgB,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,CAcjD;AAYD,wBAAgB,wBAAwB,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,CAczD"}

package/dist/lib/content.js

@@ -0,0 +1,201 @@
+export function contentToString(content) {
+    if (typeof content === 'string')
+        return content;
+    if (content == null)
+        return '';
+    if (Array.isArray(content)) {
+        return content
+            .map((b) => {
+            if (typeof b === 'string')
+                return b;
+            const block = b;
+            // OpenAI blocks carry type:'text'; Gemini-lineage agents (Qwen Code,
+            // AionUI) send part-style `{ text }` with no type at all — accept any
+            // block whose `text` is a string and whose type doesn't say it's
+            // something else. (#200)
+            if (typeof block?.text === 'string' && (block.type === 'text' || block.type === undefined)) {
+                return block.text;
+            }
+            return '';
+        })
+            .join('');
+    }
+    return '';
+}
+// Rough input-token count (~4 chars per token) used for routing and budget
+// checks: message text, assistant tool-call arguments, and the tool schemas
+// the request carries. Tool schemas go upstream on every turn and count
+// against the provider's context window and TPM ceiling. An agent client like
+// Claude Code sends ~16k tokens of them next to ~4k of messages, so leaving
+// them out routed its requests to models they could never fit (a Groq 8k-TPM
+// model answered 413 to an "estimated" 4k request that was really 17.7k).
+export function estimateInputTokens(messages, tools) {
+    const messageTokens = messages.reduce((sum, m) => {
+        const calls = (m.tool_calls ?? [])
+            .reduce((n, call) => n + call.function.name.length + call.function.arguments.length, 0);
+        return sum + Math.ceil((contentToString(m.content).length + calls) / 4);
+    }, 0);
+    const toolTokens = tools && tools.length > 0 ? Math.ceil(JSON.stringify(tools).length / 4) : 0;
+    return messageTokens + toolTokens;
+}
+export function flattenMessageContent(messages) {
+    return messages.map((m) => ({
+        ...m,
+        content: contentToString(m.content),
+    }));
+}
+// True if the content array carries an image block. OpenAI's multimodal
+// envelope uses `{ type: 'image_url', image_url: { url } }`; some clients send
+// a bare `{ type: 'image', ... }`.
+export function contentHasImage(content) {
+    if (!Array.isArray(content))
+        return false;
+    return content.some((block) => {
+        const type = block?.type;
+        return type === 'image_url' || type === 'image';
+    });
+}
+// True if any message carries an image content block. Used to route image
+// requests only to vision-capable models (#118, #125).
+export function messageHasImage(messages) {
+    return messages.some((m) => contentHasImage(m.content));
+}
+// Drop image blocks from every message so a text-only model (e.g. the fusion
+// judge, which synthesizes from panel text answers) never receives pixels.
+export function stripImagesFromMessages(messages) {
+    return messages.map((m) => {
+        if (!Array.isArray(m.content))
+            return m;
+        const kept = m.content.filter((block) => {
+            const type = block?.type;
+            return type !== 'image_url' && type !== 'image';
+        });
+        if (kept.length === m.content.length)
+            return m;
+        return { ...m, content: kept.length > 0 ? kept : '' };
+    });
+}
+// GitHub Models refuses a long history outright (413) rather than truncating
+// it the way most providers do, so an oversized body turns every github hop in
+// the fallback chain into a wasted attempt. 7500 is the input budget we trim
+// to: comfortably inside the ~8000-token ceiling observed live, with room for
+// the tool schemas and the system prompt the gateway prepends.
+export const GITHUB_MAX_INPUT_TOKENS = 7500;
+// The proxy's own chars/4 input estimate, per message — same arithmetic the
+// routing/budget paths use, so the guard below trims against the number the
+// rest of the gateway reasons about.
+function estimateMessageTokens(message) {
+    return Math.ceil(contentToString(message.content).length / 4);
+}
+// Trim one message's text down to a token budget. Non-string content
+// (multimodal envelopes) is returned untouched — dropping or rewriting image
+// blocks here would corrupt the envelope, and text is the only thing this
+// guard is allowed to shorten.
+function truncateMessageText(message, budget) {
+    if (typeof message.content !== 'string')
+        return message;
+    const maxChars = Math.max(0, budget) * 4;
+    return message.content.length <= maxChars
+        ? message
+        : { ...message, content: message.content.slice(0, maxChars) };
+}
+/**
+ * Fit a message list inside GitHub Models' input ceiling before dispatch.
+ * Keeps the leading system prompt verbatim (its instructions are the whole
+ * point of prepending one) and then as many of the NEWEST messages as the
+ * remaining budget allows, dropping the oldest turns first — the same trade a
+ * client's own context window makes. A single message that blows the budget on
+ * its own is truncated rather than dropped, so the turn is never dispatched
+ * with nothing to answer.
+ *
+ * Returns the ORIGINAL array when the request already fits, which is both the
+ * common case and the cheap one: callers can hand every github-routed request
+ * through here without copying anything.
+ */
+export function truncateMessagesForGithub(messages, budget = GITHUB_MAX_INPUT_TOKENS) {
+    if (messages.length === 0)
+        return messages;
+    const total = messages.reduce((sum, m) => sum + estimateMessageTokens(m), 0);
+    if (total <= budget)
+        return messages;
+    const systemIdx = messages.findIndex(m => m.role === 'system');
+    const systemMessage = systemIdx >= 0 ? messages[systemIdx] : null;
+    const rest = messages.filter((_, i) => i !== systemIdx);
+    const systemTokens = systemMessage ? estimateMessageTokens(systemMessage) : 0;
+    let used = systemTokens;
+    const kept = [];
+    // Newest → oldest: the most recent context is the context the answer needs.
+    for (let i = rest.length - 1; i >= 0; i--) {
+        const tokens = estimateMessageTokens(rest[i]);
+        if (used + tokens > budget)
+            break;
+        kept.unshift(rest[i]);
+        used += tokens;
+    }
+    // Nothing fit alongside the system prompt: keep the newest message in
+    // truncated form instead of sending an empty conversation.
+    if (kept.length === 0 && rest.length > 0) {
+        kept.push(truncateMessageText(rest[rest.length - 1], budget - systemTokens));
+    }
+    return systemMessage ? [systemMessage, ...kept] : kept;
+}
+// Harden the OUTBOUND envelope so strict OpenAI clients don't choke on the
+// shape variations free-tier providers emit. Complements normalizeOutboundContent
+// (which fixes array content) by fixing the *frame* fields, not the content:
+//   - `model` coerced to a string (some providers send null/number).
+//   - `choices[].finish_reason` defaulted to null when absent — the field must
+//     exist per spec; Rust agents (the #200 class of compat bugs) branch on it.
+//   - a literal `tool_calls: null` on a message/delta is DELETED — OpenAI omits
+//     the field entirely when there are no calls, and `null` breaks clients that
+//     test for its presence (this is the response-side mirror of the #200 fix).
+// Content, real tool_calls arrays, reasoning_content, and every other field are
+// left untouched. Mutates and returns the same object (frames are parsed fresh
+// per SSE line, so in-place mutation is safe), matching normalizeOutboundContent.
+export function sanitizeResponse(payload) {
+    const p = payload;
+    if (!p || typeof p !== 'object')
+        return payload;
+    if (p.model != null && typeof p.model !== 'string')
+        p.model = String(p.model);
+    if (Array.isArray(p.choices)) {
+        for (const choice of p.choices) {
+            if (!choice || typeof choice !== 'object')
+                continue;
+            const c = choice;
+            if (c.finish_reason === undefined)
+                c.finish_reason = null;
+            if (c.message && typeof c.message === 'object' && c.message.tool_calls === null)
+                delete c.message.tool_calls;
+            if (c.delta && typeof c.delta === 'object' && c.delta.tool_calls === null)
+                delete c.delta.tool_calls;
+        }
+    }
+    return payload;
+}
+// Normalize the OUTBOUND (provider → client) shape so we honor the OpenAI
+// contract on the response path the same way `contentToString` does on the
+// request path. Per spec, `choices[].delta.content` (streaming) and
+// `choices[].message.content` (non-stream) are strings; some providers
+// (e.g. Mistral magistral) return an array of content blocks. Forwarding the
+// array verbatim breaks string-consuming clients ("expected str, got list")
+// and, mid-stream, drops the turn's tool calls. We coerce array content to a
+// string while leaving `tool_calls` and every other field untouched. Mutates
+// and returns the same object (chunks are parsed fresh from JSON per frame, so
+// in-place mutation is safe). Non-array content passes through unchanged. (#166)
+export function normalizeOutboundContent(payload) {
+    const choices = payload?.choices;
+    if (!Array.isArray(choices))
+        return payload;
+    for (const choice of choices) {
+        const delta = choice?.delta;
+        if (delta && Array.isArray(delta.content)) {
+            delta.content = contentToString(delta.content);
+        }
+        const message = choice?.message;
+        if (message && Array.isArray(message.content)) {
+            message.content = contentToString(message.content);
+        }
+    }
+    return payload;
+}
+//# sourceMappingURL=content.js.map