llms.txt: the router is described, not linked (it answers POST only): POST /api/v1/route was listed as a fetchable URL and answers 404 to a GET, so a language model following it got nothing. The endpoint stays described as a plain line in the section body, and the agent manifest takes its place as the GET link. The llms.txt test now checks that every URL the file lists answers and that the POST-only endpoint is described, not linked. (#177)

ddiff --git a/server/http/discovery.js b/server/http/discovery.js
iindex 026f0b1..73ed898 100644
---- a/server/http/discovery.js
++++ b/server/http/discovery.js
@@@ -46,6 +46,27 @@ const publicPages = () => [
      { path: '/updates', changefreq: 'daily', priority: 0.5 },
  ];
  
++/**
++ * llms.txt: the shared kit's llmsTxt, with one thing it cannot say. A POST-only endpoint is not a
++ * page to fetch — a link to it is a URL that answers 404 to a GET, and a language model that
++ * follows it gets nothing. So a section may carry notes: plain lines that describe such an
++ * endpoint (POST /api/v1/route — ...) instead of linking it. Every real GET link stays a link.
++ */
++function llmsTxt({ name, summary, details, sections = [] }) {
++    const out = [`# ${name}`, '', `> ${String(summary || '').replace(/\s+/g, ' ').trim()}`, ''];
++    if (details) out.push(String(details).trim(), '');
++    for (const sec of sections) {
++        if (!sec) continue;
++        const links = (sec.links || []).map((l) => '- [' + l.title + '](' + l.url + ')' + (l.note ? ': ' + String(l.note).replace(/\s+/g, ' ').trim() : ''));
++        const notes = (sec.notes || []).map((n) => String(n).replace(/\s+/g, ' ').trim());
++        if (!links.length && !notes.length) continue;
++        out.push('## ' + sec.title, '');
++        if (links.length) out.push(...links, '');
++        if (notes.length) out.push(...notes, '');
++    }
++    return out.join('\n');
++}
++
  function createDiscoveryRoutes(ctx) {
      const { config } = ctx;
      const r = asyncRouter();
@@@ -61,7 +82,7 @@ function createDiscoveryRoutes(ctx) {
      });
  
      r.get('/llms.txt', (_req, res) => {
--        res.type('text/plain').set('Cache-Control', TEXT).send(seo.llmsTxt({
++        res.type('text/plain').set('Cache-Control', TEXT).send(llmsTxt({
              name: SITE_NAME,
              summary: 'OpenVibe.Actor: an AI agent router and OpenVibe\'s own agent. A task in plain words goes to the best agent system for it, by what the task needs and the caller\'s mode (cheapest, balanced, best, fastest, private); a second model family checks the answer; the task records its cost to a fraction of a cent and why that agent was picked.',
              details: 'Agents today: the OpenVibe runtime (a DeepSeek model acting through OpenVibe.Search and OpenVibe.Tools: DNS, mail records, certificates, headers, WHOIS and more), OpenAI\'s web agent (live web search with sources), and an open model on OpenVibe\'s own server for private mode. Coding goes to OpenVibe.Codes. Every page is server-rendered and readable without JavaScript. The API: POST /api/v1/tasks, GET /api/v1/tasks/:id, GET /api/v1/tasks/:id/events (server-sent events), POST /api/v1/tasks/:id/cancel, GET /api/v1/agents and POST /api/v1/route (public dry run). Tasks are platform.task@1 in openvibe-contracts.',
@@@ -74,11 +95,13 @@ function createDiscoveryRoutes(ctx) {
                  ] },
                  { title: 'Machine-readable', links: [
                      { title: 'The agents (JSON)', url: abs('/api/v1/agents'), note: 'actor.agent-list-result@1' },
--                    { title: 'Dry-run the router', url: abs('/api/v1/route'), note: 'POST { task, mode? } → which agent and why (platform.placement-result@1)' },
++                    { title: 'The agent manifest (JSON)', url: abs('/.well-known/agent.json'), note: 'what this site tells an agent about itself' },
                      { title: 'The free tier (JSON)', url: abs('/limits.json') },
                      { title: 'Sitemap', url: abs('/sitemap.xml') },
                      { title: 'Full text for language models', url: abs('/llms-full.txt') },
                      { title: 'Release metadata (JSON)', url: abs('/release.json') },
++                ], notes: [
++                    'POST /api/v1/route — dry-run the router: send { task, mode? } and get which agent would take it and why (platform.placement-result@1). It answers POST only, so there is no page here to fetch; the API is described at /docs.',
                  ] },
                  { title: 'Elsewhere', links: [
                      { title: 'OpenVibe.Codes', url: 'https://openvibe.codes', note: 'the coding-agent harness Actor sends coding work to' },
ddiff --git a/test/discovery.test.js b/test/discovery.test.js
iindex a812ceb..1cb5dab 100644
---- a/test/discovery.test.js
++++ b/test/discovery.test.js
@@@ -56,6 +56,26 @@ const PRIVATE = ['/tasks', '/auth/', '/api/', '/limits.json'];
              for (const p of ['/tasks', '/auth/']) assert.ok(!r.text.includes(`https://openvibe.actor${p}`), `private path in llms.txt: ${p}`);
          });
  
++        await check('llms.txt: every URL it lists answers a GET, and the POST-only endpoint is described, not linked', async () => {
++            const r = await t.get('/llms.txt');
++            const SITE = 'https://openvibe.actor';
++            // Every link the file names must answer when it is fetched as a link.
++            const links = r.text.split('](').slice(1).map((s) => s.split(')')[0]).filter((u) => u.startsWith('http'));
++            assert.ok(links.length >= 8, 'too few links: ' + links.length);
++            for (const url of links) {
++                if (!url.startsWith(SITE)) continue;   // the other OpenVibe sites are not this app to fetch
++                const res = await t.get(new URL(url).pathname);
++                assert.ok(res.status < 400, url + ' answers ' + res.status);
++            }
++            // The router answers POST only, so it is described in the section body, never linked.
++            const endpoint = '/api/v1/route';
++            assert.ok(!r.text.includes('](' + SITE + endpoint + ')'), 'the POST-only endpoint is still a link');
++            const described = r.text.split('\n').find((l) => l.startsWith('POST ' + endpoint + ' — '));
++            assert.ok(described, 'the POST-only endpoint is not described');
++            assert.ok(described.includes('POST only'), 'the description does not say it answers POST only');
++            assert.ok(!described.includes(']('), 'the described POST endpoint is a link');
++            assert.ok(!described.includes('http'), 'the described POST endpoint names a URL to fetch');
++        });
          await check('llms-full.txt: 200 text/plain, the site title, and every page the sitemap lists', async () => {
              const r = await t.get('/llms-full.txt');
              assert.strictEqual(r.status, 200);