Changelog

v0.11.5 September 15, 2026

Added

• Skill v0.30.0: new section "Before you call the site done" — a public twenty-item launch checklist, audited against Hermes Startup's own live site on 2026-09-13 (two icon gaps found, both fixed the same day). Compressed to the ten points that decide whether a site is real: per-page titles and descriptions, the single above-the-fold action, an owned 404, the full icon and Open Graph set, robots.txt and sitemap.xml, alt text and image weight, real-phone testing, privacy and terms pages, no cookie banner without cookies, and forms that survive a failed send.
• Skill v0.31.0: new section "Make the profile say what you do". A post earns the visit. The profile decides the follow, and the bio carries the link. Name real things ("Hermes Agent" over "AI agents", "Building Hermes Startup" over "building in public"), keep the cover image in step with the current headline, and check the cover on both the phone app and desktop before uploading, because X keeps only a middle strip on the app, the avatar covers the lower left, and the app's icons cover the top right. From the founder's September 2026 profile rebuild (four upload rounds; the safe zone is written down so it is never rediscovered).
• Skill v0.31.0: new section "Back up the business off the machine". A scheduled encrypted copy to off-site storage plus another copy on a second machine, a restore that is proven by decrypting and comparing, a storage guard, and the traps that make a backup incomplete while it reports success (files the job cannot read are skipped without an error; the same command behaves differently across machines and shells). The old safety net is kept until the new pipeline lands real scheduled runs.
• Skill v0.31.0: "What the founder tried that did not work" gains mistake #9 (a copy job reported success while skipping files it could not read; a sorted listing then suggested gaps that did not exist). Fix: run as the owning account, confirm readability, and prove a backup by restoring it, never by the exit code.
• Skill v0.31.0: "Continuous guidance" gains a fifth rule: own the follow-ups. A predictable next step is scheduled, never left with the user as a reminder.

Changed

• Backend upgraded to Hermes Agent v2026.9.14 (v0.21.3). The remote-desktop patch release: a Desktop wake burst can no longer replay an already-rotated refresh token and revoke the whole session, a slow identity provider no longer freezes the status endpoint, and long-lived processes stop leaking duplicate state.db writer handles. Also in this window: reasoning-effort selection on every model picker, OpenRouter OAuth login, a daily MCP re-auth nudge in Desktop, and refreshed image and video model catalogs. The Hermes Startup guided path to your first $1 runs on the newest core.

v0.11.4 September 13, 2026

Changed

• Tool/provider catalog claim updated to 2,000+ API tools from 50+ providers (the underlying catalog grew past 2,000 tools and 60+ providers; site, llms, skill, and open-core surfaces updated in lockstep).

Added

• Skill v0.29.0: new founder mistake #8 (publishing the price without the tool's input contract). Buyers could see what a tool cost but not how to call it, so every first call was trial and error — and a rejection that ran nothing looked like an uncertain paid outcome instead of a clean refusal. The provider's own inspection response already carried the tool's input schema and method; the storefront discarded both. Fix: relay the tool's real input contract wherever you publish the offer (bounded and validated), and treat a rejection that ran nothing as a clean no-charge refusal — holds are only for outcomes you cannot see.

v0.11.3 September 12, 2026

Added

• Skill v0.28.0: the bot-team section now notes that Bot Mode runs a whole roster without a separate backend per profile or per tick, so several bots fit on an ordinary laptop.
• New entry in "What the founder tried that did not work": the release pipeline that mutated state (version, changelog, release tag) before it could verify anything, and the fix the founder now runs, which is to log every step, refuse a second bump while a release is unverified, and record the tag only after the live checks pass.

v0.11.2 September 12, 2026

Added

• Skill v0.27.0, captured from the v2026.9.11 platform review. "Start on free inference": free inference and connectors arrive with one sign-in, so the audit, the ten ranked ideas, and the first research pass cost nothing, and the US$10 appears only when the user picks a direction. Be straight about the ceiling of a free allowance before a job runs past it. "Sign in and fill forms without exposing a secret": the password-blind credential vault (1Password, Bitwarden, or the local vault) lets the agent sign up, pay, and fill addresses without ever holding the secret; two-factor codes stay with the user, and every account and its terms remain the user's approval. "If something looks wrong, check the store before rebuilding": run the built-in doctor check first, because damaged data is quarantined rather than overwritten and unprovable repairs are refused. Also: prefer the commit-pinned plugin catalog for plugin installs, and give each business its own profile now that profiles keep their own allow-lists, credentials, and vault secrets.

Fixed

• Stale free-tier line in the paid-user skill: the marketing section still described a three-idea free tier while the product gives all ten ideas free, researched and ranked. Both mentions now match the free-tier contract.
• The share card for the finished-questions milestone still read "Three ideas built just for me." while the free tier returns all ten ideas. It now reads "All ten ideas, built for me." The longer approved wording wrapped to two lines at the card's display size, so it was shortened to fit; the card CTA still carries "start free".
• The release pipeline can no longer ship a silent partial release. deploy_site.sh used to bump VERSION, promote the changelog and record the release tag before it had deployed anything, so a run that died mid-flight left a bumped version and a tag with nothing live (and the next run could only refuse). It now logs every step to a deploy evidence file with a fail-loud trap, refuses to start a second bump while an unverified release marker exists, defers the tag until the live checks pass, and clears the marker only after they do. bump_version.sh gained the deferred-tag mode the deploy now uses.

v0.11.1 September 12, 2026

Added

• Skill v0.26.0: new founder mistake #6 (trusting one cost shape across the whole catalog). The first paid run settled cleanly and the settlement reader trusted that shape everywhere; other endpoints from the same provider report the billed cost differently (integer micro-units instead of a dollar number), so paid runs on them failed after the provider call. Fix: capture the raw cost field each endpoint actually returns before you resell it, settle only from shapes you have captured, and fail closed and hold for reconciliation when nothing verifiable appears or two shapes disagree.

Changed

• Backend upgraded to Hermes Agent v2026.9.11 (v0.21.2). The session-store patch release: one corrupt row or a damaged search index no longer takes down the session list, exports, or the conversation itself; sessions can never read or bind to another profile's store; and opening the store behind a busy gateway no longer stalls (4-20 s before, 0.01 s now). Also in this window: multi-profile isolation hardening (allow-lists, credentials, and vault secrets stay on the profile they belong to), desktop Bot Mode backends no longer spawn per profile, a password-blind credential vault (1Password, Bitwarden, or the local vault) lets the agent sign in and fill forms without ever seeing a secret, a curated SHA-pinned plugin catalog, and free inference with connectors on first launch. The Hermes Startup guided path to your first $1 runs on the newest core.

v0.11.0 September 11, 2026

Added

• The agent API now publishes a machine-readable payment contract: a public offer catalog at /v1/x402 and the standard PAYMENT-REQUIRED / PAYMENT-SIGNATURE / PAYMENT-RESPONSE headers, so an agent can read the terms of the prepaid balance and authorize a spend without a human relaying numbers. The terms are the ones that already applied: one prepaid balance, no subscription, and a ceiling set when the spend is authorized. **Advertised, not required** — with enforcement off, calls behave exactly as they did before, and a payment header never substitutes for an account token or widens a spend. Agent-facing surfaces (for-agents.md, llms.txt, index.json) now link the catalog.

v0.10.3 September 9, 2026

Changed

• Reframe the paid continuation panel from cheap-labour framing to stake framing: "A one-time US$10. That's the whole payroll." is now "A one-time US$10. Your stake in a business your agent runs." The balance is fuel for a business the agent runs, not the price of cheap labour; keeps every honesty line (no subscription, balance never expires, approval on every spend) intact.
• Tighten the step-2 claim from "No setup required = Faster results" to "No setup required. Your first results come faster." (claims results specifically, not a vaguer speed promise).
• llms.txt opener now mirrors the intent-first for-agents.md opener ("Use it when a user wants proven ideas matched to them...") instead of leading with mechanism.
• Move the "AI business builders" link out of the site header (footer only) and make the header wrap instead of clipping on narrow screens.
• Fix the ai-business-builders hero lead: the long "No sponsored slots..." line was clipped on desktop by the nowrap rule, letting it slide under the activation card. It now wraps within the column.
• Collapse the seven competitor entries on ai-business-builders into expandable one-line rows (full text stays in the DOM for crawlers and LLMs; humans get the grid + Hermes Startup entry + FAQ without a 3,000-word scroll).

v0.10.2 September 8, 2026

Added

• New "AI business builders compared" page (/ai-business-builders): an honest, sourced comparison of eight builders in this category (Crevio, Durable, Lindy, B12, Polsia, n8n, Make, and Hermes Startup listed last and clearly marked) on what each actually builds, who it fits, the pricing shape, and where it stops, with no scores and no sponsored slots. Covered call: AI answer engines keep citing vendor comparison pages in this category, so this is our version of the format without the self-dealing. Ships with a machine-readable markdown mirror, llms.txt and sitemap entries, and homepage nav links.

v0.10.1 September 8, 2026

Added

• Skill v0.25.0: "Re-frame the sentence before you rebuild" section. When the pitch gets no response, test the sentence before touching the product. Adopted from the founder's September 2026 homepage re-frame (capability pitch to hire pitch, "Your first employee is a Hermes Agent") after the Gitroom-to-Postiz story showed the same product under a different sentence getting a completely different reception. Method: research what is performing right now (the founder's trend scan cost about $0.29 in fees), diagnose your sentence against the evidence, draft variants where only the sentence changes, preview and approve before shipping. Also adds founder mistake #5: a release was nearly called broken because the verification check did not follow redirects (fix: follow redirects and read the final page before calling anything broken).
• Skill v0.24.0: "Land your first gig: proposals that never lie" guided recipe. When a user's chosen direction is freelancing, gather the real job posting and the user's real background before drafting; any gap becomes a question, never an invented credential. Proposal structure (client's problem restated, concrete connections, one real proof point, concrete next step), same honesty in the platform profile, and exact-content approval before the user sends anything. Freelancing is the most common first-dollar path for non-technical users.

Changed

• Backend upgraded to Hermes Agent v2026.9.7 (v0.21.1). Patch rollup since v2026.8.31: cron and delivery reliability fixes (scheduled jobs start under systemd, cron messages and local DMs reach an open Desktop Bot Chat), subagent delegation fixes (pinned children keep their own provider fallback chain), approval hardening (paths and launchers can no longer bypass explicit deny rules), MCP authorization improvements, desktop session controls, browser annotations, provider and model updates, and faster file operations and startup. The Hermes Startup guided path to your first $1 runs on the newest core.

v0.10.0 September 5, 2026

Added

• Free workflows page (site/workflows.html, plus agent-readable site/workflows.md and site/workflows.js): eight ready-to-paste Hermes Agent workflow prompts, each in its own section with a per-prompt Copy button and a Copy-all-8 pack button. Prompts are original, written for Hermes Agent, and grounded in the owner's working skills; the three adopted external lessons are baked in (verify decision-critical claims; prove before authoring plus non-trigger testing; retune watches on misses and false positives). Free top-of-funnel: no signup, no subscription, and the bottom CTA leads into the free ideas path. Inspiration credited in the footer to @aiedge_'s 8-workflows thread.

Changed

• Discovery wiring for the new page: homepage footer nav now links Workflows; workflows.md is declared as a markdown alternate in the homepage head; llms.txt and index.json point agents at the page; sitemap.xml lists /workflows and /workflows.md.

v0.9.0 September 4, 2026

Changed

• Homepage re-framed to hire-frame copy (owner-approved via preview): headline now "Your first employee is a Hermes Agent"; price panel "A one-time US$10. That's the whole payroll."; tools section "Your agent brings its own tools."; team section "One hire does the work of several." — lead label, CTAs, pricing, honesty lines, FAQ, and footer tagline ("Make your first $1 online with your Hermes Agent") unchanged. Pipeline guards and copy tests updated to the new strings.

Added

• Skill v0.23.0: "Sell to agents in one command" section plus the bundled scripts/sell_to_agents.py scaffolder. One command (python3 scripts/sell_to_agents.py new ./my-store) generates a complete Agentic Commerce Protocol (ACP, agenticcommerce.dev, the OpenAI + Stripe open checkout standard) merchant endpoint for a Hermes-built store: create / retrieve / update / complete / cancel checkout sessions, authoritative server-side totals from products.json, Idempotency-Key enforcement, fail-closed payment (completing without a configured restricted Stripe key is refused 503), Stripe hosted Checkout on completion, an agent-readable for-agents.md contract, and a self-test (node --test checkout_sessions.test.mjs) that must pass before deploy. Same pattern Hermes Startup runs; honest framing: an ACP endpoint removes friction for agents already trying to buy, it does not create demand, and each agent platform runs its own merchant-listing application.

v0.8.9 September 3, 2026

Added

• Design contract: added site/DESIGN.md — the owner-approved black-and-gold design state (v0.8.8) encoded as Google's open DESIGN.md spec (colors, typography, spacing, radius, components, do's/don'ts), linted clean with @google/design.md (0 errors, WCAG AA passes on gold/ink), and linked from llms.txt so agents can read the design contract directly.

v0.8.8 September 3, 2026

Changed

• Copy: added the cost-of-inaction stakes line to the US$10 continuation panel ("Every month you wait, your Hermes Agent keeps earning you nothing. It has the tools. It just does not have the business.") and reworked the for-agents.md opener to lead with the agent's intent match (when to use) before the mechanism, preserving the outcome-claim boundary. Adopted from the Leon Abboud marketing-foundations framework (desire-based messaging, grunt test).

v0.8.7 September 2, 2026

Changed

• Design polish pass (fresh-context design-critic loop, Lenny's playbook): refined the bot-card process connectors, added a subtle left-to-right brightness ramp so the five team cards read positionally, gave the prompt card breathing room (unified inner padding), softened the hero glow, lifted badge/fine-print contrast, unified the footer seam with the gold grammar, and opened up FAQ rhythm. Aesthetics-only — no copy or structure changes.

v0.8.6 September 1, 2026

Changed

• Backend upgraded to Hermes Agent v2026.8.31 (v0.21.0, "The Pantheon Release") — ~2,475 merged PRs since v2026.8.3: cron jobs gained persistent memory and continuity (scheduled agents learn between runs, monitor jobs skip the LLM entirely when nothing changed), subagents can be steered live mid-flight with JSON-schema output validation, MCP servers became a command center with background health checks and per-server cost visibility, the agent can drive the desktop's own browser, and protected instruction files (AGENTS.md, skills, memory stores) now always require write approval alongside a deeper secret-redaction sweep. Six new model providers and per-model overrides landed as well. The Hermes Startup guided path to your first $1 now runs on the newest core, with stronger reliability and security across the board.

v0.8.5 August 31, 2026

Added

• Skill v0.22.0: "All 10 ideas free" offer change. The free tier now returns all ten researched ideas, each explained in detail and ranked with the reason why; the US$10 prepaid continuation builds and runs the chosen business (research, build, launch, and the team that keeps it running), with every external action approval-gated. Completes the ideas pool from 4 to 10 defined directions (adds niche newsletter, local lead list, template pack, micro tool, video edit service, paid cohort). Website copy, agent-discovery surfaces (index.json, llms.txt, llms-full.txt, for-agents.md, about, ai-catalog, ai-plugin, agent-skills index), homepage, and FAQ all migrated from the 3-of-10 contract. Adopted from the founder's 2026-08-30 feedback (10 free ideas, sell the build/run).
• Skill v0.21.0: "Add capabilities with community skills" section. Vetting flow for third-party skills: inspect before install (hermes skills inspect <url>), trust labels, license check before commercial bundling (MIT/Apache-2.0 with attribution), redundancy check against built-in capabilities (imagegen vs the native image tool), hub/URL installs, and taps for curated packs. Adopted from the founder's 2026-08-30 evaluation of two skills from a viral post (img2threejs, imagegen) for use with Hermes Agent and Hermes Startup.
• Skill v0.21.0: "Prove the account is yours" section. One-time challenge token placed at a URL only the owner can publish to (or posted from the official account) as the only proof a stranger cannot fake; stable verify path (/.well-known/founder-verify.txt); local machine evidence is corroboration, not proof. Adopted from the founder's own identity verification deployed 2026-08-30 on hermesstartup.com.

v0.8.4 August 30, 2026

Fixed

• Copy-event worker CORS: allow the www.hermesstartup.com origin and send Access-Control-Allow-Credentials for the visit beacon (sendBeacon preflight was blocked on fresh page loads, logging CORS errors).

Changed

• Deploy pipeline inlines the minified stylesheet into the homepage (removes the render-blocking CSS request; FCP/LCP win). Readable sources unchanged; tests and live-verify assertions updated.
• Fonts: subset Inter (inter-400.ttf 324 KB → 62 KB, inter-var.woff2 99 KB → 38 KB); retired inter-500/600/700.
• Security headers hardened (_headers): CSP extended for Cloudflare Insights + trusted types, COOP same-origin, HSTS includeSubDomains + preload, no-transform on / and /*.html.
• Site skill bundle rebuilt from reviewed sources (skill v0.20.0 "Own the outcome, not the platform").

v0.8.3 August 30, 2026

Added

• Skill v0.20.0: "Own the outcome, not the platform" section plus founder principle #6: the product is the solved problem, never the machinery; offer written as the finished job with a standard and verification, done-for-you services rebuilt with AI (YC thesis), engine as margin not message, owning the error as the trust move, Hermes Startup itself selling a verified first dollar rather than capabilities. Adopted from @yamanzdh (reported $500k in two months after clients stopped being shown the platform; claim attributed, not guaranteed) and the linked Y Combinator Startup School thesis on AI-rebuilt services businesses.

v0.8.2 August 30, 2026

Added

• Skill v0.19.0: "Check if AI assistants can find your site" section: AI-crawler robots.txt check (OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot/SearchBot, Googlebot, Bingbot), reachability, citation check on buyer-intent prompts, citable-content checks, the GPTBot/Google-Extended traps, ranked fixes with how-to-know-it-worked. Adopted from the Claude-side ai-search-visibility-audit + llm-crawler-access-check pair (davepoon/buildwithclaude #296, MIT) via the 2026-08-30 ecosystem-radar recommendation. Internal twin skill: hermes-startup-ai-visibility-audit (own-site weekly cron + monitor).

v0.8.1 August 30, 2026

Changed

• Pricing voice across the site and skill runtime (owner-directed 2026-08-30): paid version now speaks "Simple, transparent pricing. Pay as you go. Add funds, use the API, and auto top-up keeps you running. No subscriptions. Balance never expires. One-time." Homepage (hero, price panel, benefits, team note, FAQ x3, checkout success), developers.md, auth.md, contact.md, terms (md+html), and the conversation-facing continuation preview (radar/startup.py billing/payment/boundary) updated; "no automatic top-up" removed everywhere, auto top-up framed as optional and user-controlled.

v0.8.0 August 30, 2026

Added

• Skill v0.18.0: "Hire agents like employees" section: charters (owns / good output / stop line), five-field briefs (outcome, sources, constraints, deliverable, review point), one bot per owned outcome, evidence receipts ("show the tape"), staged trust ladder, weekly roster self-review. Adopted from the 2026-08-29 @coinbureau Grok Bot one-person-company thread teardown (8 X Articles by milesdeutscher, mattyp, sairahul1, N01ennn, Argona0x, ScottyBeamIO, adiix_official).

v0.7.3 August 29, 2026

Changed

• Homepage: the "A team of bots operates your business for you." section now sits above the FAQ (FAQ remains above the bottom CTA). Pure reorder, copy unchanged.
• HTML responses now send Cache-Control: no-transform, which stops Cloudflare's JavaScript Detections injection (it was blocked by our strict CSP and logged a console error; Cloudflare docs confirm no-transform suppresses the injection). Best Practices console-error audit is now fully clean.

v0.7.2 August 29, 2026

Changed

• Inter webfont subset to the site's exact character set (99.4 → 37.9 KiB woff2, ~62% smaller; wght axis 400–700 and all glyphs the site uses verified in-browser). Removed three unused static TTFs. Faster font swap on slow connections.

v0.7.1 August 29, 2026

Changed

• PageSpeed pass: the homepage stylesheet is inlined at build time (no render-blocking CSS request; mobile Performance 94 → 98+ with LCP 2.6s → 2.4s).
• Hardened security headers: Cloudflare Web Analytics beacon allowed by CSP, Trusted Types enforced, Cross-Origin-Opener-Policy: same-origin, HSTS now includeSubDomains; preload.
• Telemetry CORS: the copy-metrics Worker now accepts the www.hermesstartup.com origin, so visit/copy/metrics signals from the www homepage reach the reviewed anonymous counter (previously blocked with 403).

Fixed

• Homepage console errors removed: CSP no longer blocks the Cloudflare beacon, and /v1/visits + /v1/copy-metrics preflights no longer fail CORS from the www origin.

v0.7.0 August 29, 2026

Added

• Homepage FAQ section directly below the US$10 price panel: six native accordion items answering the objections that decide the purchase (subscription status, what the payment buys, provider setup, answer privacy, balance expiry, Hermes Agent requirement). Zero JavaScript, reduced-motion safe, and styled in the same gold reveal language as the rest of the page.

Changed

• Anti-vibe pass on the homepage (design-token playbook): the violet accent is retired from the hero headline and background gradient layers, leaving a strict black-and-gold palette; the "Step 1 / Step 2" prefixes and the 01-05 team-card numbering are removed; the background gradient stack is consolidated from six layers to four. The full-page cursor glow was restored at the owner's request after the audit flagged it.

v0.6.5 August 28, 2026

Added

• Agent-commerce discovery surface: UCP business profile at /.well-known/ucp (Universal Commerce Protocol; declares the REST service and the prepaid-wallet payment handler), a programmatic pay/use contract in for-agents.md (activate → purchase → wallet → status → gated tool run), an OpenAPI entry, and index.json / llms.txt pointers. The status and tool-selection/run API routes ship fail-closed behind an operator switch and join the public worker only after the pending adversarial review (HUMAN_GATES gate 4). See AGENT_COMMERCE.md.
• Paid-user skill v0.17.0: "Make your product payable by agents" — the machine-readable payment pattern (UCP profile, plain-markdown pay/use contract, fail-closed gated execution with reserve-then-settle caps) so users running agent-first products never start from scratch.

v0.6.4 August 28, 2026

Added

• Paid-user skill v0.16.0: paid continuation framed as a blueprint, not a plan — one complete business (idea, build, launch, and the team that runs it) ready to deploy, adopted from Whop Business Blueprints (2026-08-27).
• Paid-user skill v0.15.1: "Run your bot fleet like a garden" section — fleet upkeep rules distilled from the agent-operator playbook (plan first, three-layer instructions, silence until a real hit, draft-then-approve, outcome ownership, and one bot that fixes bots).

Changed

• open-core self-hosted variant synced to the same release: v0.15.1 with the new garden section, the fee-math rail decision, and current messaging doctrine; self-hosted semantics preserved (no hosted purchase gateway, test-mode demo only, per test_launch_package enforcement).

v0.6.3 August 25, 2026

Changed

• Performance (PageSpeed round 4): Inter now ships as a single variable woff2 (99 KB, one request, one swap moment — was four 45 KB statics; visually identical, verified side-by-side; full TTFs remain as fallback), and the aurora background fades are shortened (1.4 s → 0.6 s) so the viewport settles sooner.

v0.6.2 August 25, 2026

Fixed

• Font URLs are now version-stamped like CSS/JS (fonts were served with year-long immutable caching but unstamped URLs, so the CDN kept serving previous releases' fonts; bump_version.sh now rotates the stamp in the @font-face src list and the homepage preloads on every release).

v0.6.1 August 25, 2026

Changed

• Performance (PageSpeed round 3): Inter woff2 subsets tightened to the glyphs the site actually renders (181 KB total across all weights, down from 289 KB; full TTFs remain as fallback), the headline entrance is now transform-only so the h1 is fully opaque at first paint and stays the LCP element, and all four Inter weights are preloaded (they are all used above the fold).

v0.6.0 August 25, 2026

Added

• Paid-user skill v0.15.0: "Write copy that does its job" section — the four editing passes (boring-version test, transplant test, style guide with counterexamples, detector pass) plus structure rules, adopted from a16z crypto editor Steph Bzinn's public X article on the habits of AI writing (2026-08-24).
• Paid-user skill v0.14.0: "Make money with an X content system" section — the two-play angle engine (repackage proven attention, or create a fresh angle from own material), the candidate contract, and the winner-log feedback loop, adopted from a public 69.8M-impression guide (2026-08-25) whose author confirms the same loop ran on Hermes Agent first.
• Ideas pool: fourth defined direction "Run an agent-powered X content system" in the opportunity map. Free tier still shows 3 of 10 (recommended one first, then pool order); the paid continuation grades and ranks the full pool including the new direction.
• Homepage "your team" section (v0.6.0): "A team of bots operates your business for you", powered by Hermes Agent Bot Mode — a five-step roster (Idea Scout, Market Check, Builder, Launcher, War Room) with animated hand-off arrows, benefit-led copy matching the shipped bot-team skill (free 3 of 10 ideas, US$10 prepaid balance, approval-gated execution, local-only answers), text-wrap balance/pretty to prevent orphan lines, and prefers-reduced-motion support for the arrow animation.

v0.5.9 August 23, 2026

Changed

• Performance (PageSpeed round 2): inter-400 is preloaded alongside inter-600 (faster first paint on slow connections), the headline word-split JS was removed (the h1 is now a single CSS entrance — no DOM mutation at load), and ~10 dead CSS rule blocks were pruned (smaller stylesheet, faster parse).

v0.5.8 August 23, 2026

Changed

• Performance (PageSpeed audit pass): the hero headline now paints immediately instead of waiting on its word-by-word reveal (removes the ~2.3 s LCP render delay), Inter fonts ship as subsetted woff2 (1.27 MB → 289 KB on disk), scripts load with defer, CSS/JS are minified at deploy time, and versioned CSS/JS/fonts now cache immutably (stamps bump every release).

Added

• Skill v0.12.0: paid-user guidance now opens with the founder's working principles (money first, test demand before building, hard-to-copy value, evidence over opinion, one next action), a hard-to-copy value checklist, a start-to-finish path map (name the buyer first, honest bounded offer, get paid with a live rehearsal, one verified channel, first customer, scale only with evidence), and a marketing, distribution, and customers doctrine (research before writing, one quality bar, free tier as funnel, agent-first surfaces, listening through the product, never promise revenue), and closes with an honest "what the founder tried that did not work" section (staggered automation against rate limits, verified fallbacks for dead capabilities, live-test over stale notes, failure-messaging checks) so users skip the founder's mistakes.

v0.5.7 August 23, 2026

Added

• Feedback: the in-session agent now asks one natural question at a moment that matters (after a win, a fix, a frustration), sends anonymous feedback cards to the founder, and follows a full feedback protocol. Consent is asked once after the free ideas; "stop feedback" ends it any time.

v0.5.6 August 22, 2026

Fixed

• Hero Step 1 / Step 2 lines no longer wrap onto two lines on wide screens (removed conflicting text-wrap: balance on .hero-lead-line).

Updated

• US$10 price card now responds on hover: subtle lift, warmer gold border, deeper shadow (matches the site's existing micro-interactions).

v0.5.4 August 22, 2026

Updated

• Footer tagline ("Make your first $1 online with your Hermes Agent.") and the Privacy/Terms links are now right-aligned.

v0.5.3 August 22, 2026

Updated

• Homepage copy de-duplicated: each promise now appears once, phrased differently ("3 of your 10 ideas, free", "3 ideas, each explained in detail, at no cost", "ideas with demand that already exists", "No category to invent", footer tagline "Make your first $1 online with your Hermes Agent").
• Hero Step 1 reads "Answer 10 questions. Get 3 proven ideas, matched to you" and stays on one line on desktop; activation prompt now says "proven business ideas that fit me".
• Paid section now leads with a premium US$10 price card: "One-time. Hermes Startup builds, launches and runs your business for you."

v0.5.2 August 22, 2026

Updated

• Homepage and About copy now lead with proven demand instead of novelty: "unique ideas" is replaced by "proven business ideas, matched to you" (hero, free-path, activation prompt, meta/OG/Twitter descriptions, JSON-LD, About), and the site now states that every idea is a business people already search for and pay for (approved via preview 2026-08-22).
• In-agent paid-continuation copy synced to the approved homepage copy across all skill bundles (radar payload, SKILL.md continuation script, adapter tests, and the about page): the offer headline is now "A money-making business, built for you." and the billing line reads "Prepaid balance. No subscriptions. No automatic top-up." — the "One-time" qualifier was dropped everywhere, including the offer boundary statement. Factual descriptions ("US$10 funds the API calls used to start building…") and historical changelog entries are untouched.

v0.5.1 August 22, 2026

Updated

• Backend upgraded to Hermes Agent v2026.8.19 (v0.20.5) — a patch release rolling up ~323 merged PRs since v2026.8.18: Bot Mode group-room threads, foldable conversation summaries, PDF/file attachments with drag & drop, the keyless web tier (5-vendor free rotation with ring failover, web search on fresh installs with zero keys), CLI polish (fuzzy /model picker, Ctrl+P command palette, richer /status), execution-discipline and runtime stall guards from the Composio eval findings, hermes update receipts and fleet --plan verification, hermes worktree list/prune, the opencode-free zero-auth provider, multi-question clarify, desktop performance work, and cron jobs gaining persistent memory and per-job reasoning effort.

v0.5.0 August 22, 2026

Added

• Agent discovery round 2: Content-Signal preferences (ai-train=no, search=yes, ai-input=no) in robots.txt and as a response header; RFC 8288 Link headers (api-catalog, service-desc, service-doc, describedby); RFC 9727 API catalog at /.well-known/api-catalog; Agent Skills discovery index at /.well-known/agent-skills/index.json (sha256 digest kept in sync by the build); ARD AI Catalog manifest at /.well-known/ai-catalog.json; auth.md statement; deterministic Accept: text/markdown content negotiation via a Pages Function with static markdown mirrors for every page (about.md, contact.md, developers.md, privacy.md, terms.md, changelog.md); RFC 9218 rate-limit headers (RateLimit-* + Retry-After) on the api.hermesstartup.com copy-metrics worker.

Changed

• Premium pass (founder-approved on the throwaway preview): headline back to "Hermes Startup helps you make money with your Hermes Agent"; hero lead now two labeled steps (1 Free: "Answer personal questions. Get 3 business ideas unique to you"; 2 Optional: "Pay Hermes Startup US$10 once. It builds and runs your idea", one line at every viewport >= 1246px); continuation heading back to "A money-making business, built for you."; "After launch, it keeps working" bullet restored; 12 hand-drawn SVG icons replace the unicode glyph chips; soft gold cursor glow (desktop hover only, hidden under reduced motion); page-wide ambient gold wash; hero gap/lead clamps and text-wrap:balance keep both steps on one line; Inter fonts shipped with the site. Hero numerals read "Step 1" / "Step 2" with a slightly smaller gold prefix (0.9em) so the step labels stay explicit and both lines still fit at every viewport >= 1246px.
• Site-wide messaging sync (founder-approved, 2026-08-22): title, meta description, og/twitter tags, JSON-LD, page footers, about pages, and every agent surface (llms.txt, llms-full.txt, for-agents.md, index.json, ai-plugin.json, ai-catalog.json, agent-skills index, skill bundle) now lead with "make money with your Hermes Agent"; "Answer personal questions. Get 3 business ideas unique to you." as the discovery line; og-image social card regenerated. "Make your first $1" remains brand essence only, kept as the milestone inside process descriptions.
• Footer alignment (founder-approved, 2026-08-22): tagline and contact row now left-aligned at every viewport (was right-aligned on desktop), so the whole footer reads left to right.

v0.4.0 August 22, 2026

Added

• Agent-readiness release: /about, /contact, and /developers pages (trust anchors + developer portal, 500+ chars each); expanded OpenAPI 3.1 description covering every public resource with operationIds, typed schemas, JSON error contract, and rate-limit conventions, with a JSON mirror at /openapi.json; structured JSON 404 for unknown /api/* paths plus /api/openapi.yaml|json aliases; robots.txt allowlist for DeepSeekBot, Applebot-Extended, Bytespider, Meta-ExternalAgent, OAI-SearchBot, cohere-ai, and CCBot; Vary: Accept, Accept-Encoding on all machine-readable responses; Organization + offers in the homepage JSON-LD identity graph; explicit when-to-use guidance in llms.txt, llms-full.txt, for-agents.md, and index.json; canonical extensionless sitemap with the new pages.

Changed

• Outcome-first site copy (owner-approved via preview rounds, 2026-08-22): headline now leads with the money-making outcome and names the product specifically ("Hermes Startup helps you make money with your Hermes Agent"; never the vague word "AI"); hero line "US$10 funds the tools for Hermes Startup to build a money-making business for you"; "A money-making business, built for you." section heading; footer, meta/og/twitter tags, JSON-LD, and agent-facing surfaces (llms.txt, llms-full.txt, for-agents.md, index.json, ai-plugin.json, site/skill bundle) updated to the same angle. "Make your first $1" remains brand essence only. Social card (og-image.png) regenerated with the new headline.
• Visual: the gold glow is now one continuous fixed page-level layer plus one cursor-reactive layer, so it flows seamlessly across every black surface with no section box outlines, corner artifacts, or divider seams; hero lead paragraph wraps cleanly on desktop and the money line renders as the gold hero line.

Added

• Product skill (site/skill/startup) v0.11.0: bundled curated idea-signal pool (scripts/radar/pool/app-replacement-signals.json, 38 vetted app-replacement signals with verdict, category, and advertised price) as a second creative hypothesis source during the opportunity audit, alongside the existing public-posts source. Pool entries are labeled unverified market facts: the agent verifies current public pricing on the product's own site before any claim, keeps the pool private to the user, writes original idea copy, and cross-checks against the user's assets and an independent market check before recommending. The pool is a founder-refreshed snapshot; installed copies are point-in-time lists.

v0.3.0 August 18, 2026

Added

• Product skill (hermes-skill/startup) v0.10.0: new "Turn any website into an API" guidance (loads the official website-to-api guide in-flow and drives it end to end when a site has no public API but loads JSON) and a new "Continuous guidance" standing rule (after every deliverable, Hermes Startup always suggests the specific optimal next step; the user never has to ask what comes next).

Removed

• "New: Website to API" homepage section removed (owner-approved via staging preview 2026-08-18). The capability stays in the product: the guide (/website-to-api.md) remains live and referenced in agent-facing surfaces, and the product skill now loads it in-flow for paid users when a site has no public API.

Changed

• Visual polish (owner-approved via staging preview): header is transparent at the top of the page and becomes a clean near-solid bar once scrolled; body background flattened to one dark canvas; the warm glow band above the footer removed so header and footer blend smoothly with the page.

v0.2.1 August 18, 2026

Removed

• Bot Team customer surface removed from the public site at the owner's request (2026-08-18): homepage "your team" section, the one-paste prompt change, and the continuation-heading change reverted; agent-facing surfaces (llms.txt, llms-full.txt, for-agents.md, index.json) returned to their pre-bot-team state; the public /bot-team/skill/ bundle and its sync/parity machinery removed. The bot-team feature itself stays under canary in the repo (bot-team/) with no customer-facing claim. Shipping process corrected: any future site copy goes through a staging preview and explicit owner approval before deploy.

v0.2.0 August 18, 2026

Added

• Bot Team customer surface shipped (owner-approved 2026-08-17): homepage one-paste prompt now installs the bot-team skill (https://hermesstartup.com/bot-team/skill/SKILL.md) and the new "your team" section introduces Idea Scout, Market Check, Builder, Launcher, and the War Room. The bot-team skill bundle (bot-team/skill/) is served byte-identically from /bot-team/skill/ (synced by scripts/sync_bot_team_skill.py in the deploy path).

v0.1.20 August 18, 2026

Added

• Bot Team live feature: bot-team/ carries the customer bot templates (Idea Scout, Market Check, Builder, Launcher), the War Room guide, the one-paste start prompt, the provisioning skill (hermes-startup-bot-team v0.1.0), the balance-gate rules, the loopback mock-wallet E2E harness, and BOT_TEAM_EVIDENCE.md (2026-08-17 end-to-end journey on a throwaway persona: interview → free 3 ideas → mock US$10 balance → provisioning → paid build/launch jobs → low-balance refusal, all verified). Seller-side: Content and Support bots completed (models, canonical Bot Chats, roster), weekly War Room routine live in the base cron store delivering to the owner's Telegram. Customer-facing claim pending owner approval (bot-team/site-copy-draft.md).
• Self-hosted open-core skill updated to v0.7.0 with the same founder-built guidance set as the hosted v0.9.0 skill: "Getting paid: Stripe setup (API and MCP)" including the prepaid-balance pattern, payment alerts, the silent-on-success uptime watchdog, the verified X selling doctrine, the release-to-post loop, the bot-team pattern, and saving every repeated workflow as a skill. The self-hosted variant stays distinct: no hosted payment service, test-mode demo only.

v0.1.19 August 17, 2026

Added

• Paid-user skill v0.8.0 now includes "Getting paid: Stripe setup (API and MCP)": full fast-path guidance to connect a user's own Stripe account — official Stripe MCP (OAuth, test-first, live switch) and Stripe API (restricted key, hosted Checkout, signed webhook, live gate).
• Paid-user skill v0.9.0 adds founder-built operation patterns: payment alerts (know the moment a live payment lands), a silent-on-success uptime watchdog (5-minute checks, messages only on failure and recovery), the verified X selling doctrine (copy-link share about 40x a like, replies about 10x, report the worst signal, mutual-follow original posts boosted, about 80-hour relevance window, from X's open-sourced ranking code checked 2026-08-13), the release-to-post loop (each Hermes Agent release becomes one plain-language post), the bot-team pattern (one bot per job, cheap model for watching, strong model for writing), and saving every repeated workflow as a skill.
• The getting-paid section gains the prepaid-balance pattern: collect once, debit per job, low-balance alert, manual top-ups, opt-in auto-top-up only.

v0.1.18 August 17, 2026

Updated

• Backend upgraded to Hermes Agent v2026.8.16 (v0.20.2) — a patch release rolling up ~397 merged PRs since v2026.8.13: desktop multi-gateway Connections registry, background MCP health checks with re-auth nudges, hermes:// deep links to install MCP servers, MCP fleet cost/usage overlay, gateway persisted model routes and /loop completion, Telegram DM topics, prompt caching for LiteLLM Claude on the OpenAI wire, cron hardening, profile-scoped auth resolution, and installer robustness on Linux and Windows.

v0.1.17 August 16, 2026

Added

• Site-wide "gold reveal" hover language: every interactive box now wakes up the same way the hero activation card does — a gold border, a warm gold bloom, a gentle lift, and a warm paper tint on the light tiles (provider grid, capability chips, and footer contact chips). Pointer-only, no-JS safe, and fully collapsed under reduced-motion preferences.
• Mobile premium beat: a one-shot gold sweep across interactive boxes and a subtle tap-press on touch, so the same warmth reads on phones without hover.

v0.1.16 August 16, 2026

Added

• New service: turn any website into an API for your agent. Hermes Agent records one browser session on a site, distills the site's hidden JSON calls into a static API client (official Hermes skill har-derived-api-client), and calls them directly with plain HTTP from then on. No browser in the loop for every request. Public guide: https://hermesstartup.com/website-to-api.md, linked from the homepage, llms.txt, the agent handoff, and the structured index.
• Verified example: recorded one Wikipedia search, derived the private endpoint w/rest.php/v1/search/title with its q and limit parameters, and replayed it with a plain HTTP request — same pages, no browser (verified 2026-08-16).

v0.1.15 August 16, 2026

Added

• The US$10 paid continuation is now live end-to-end: a customer's Hermes Agent calls the purchase gateway, receives a real Stripe Checkout URL, and the paid balance is credited after a successful payment. The public skill was upgraded (v0.6.1 → v0.7.0) with the exact purchase-gateway contract, replacing the earlier "not currently purchasable" step. Verified with two real live US$10 payments (2026-08-16).

v0.1.14 August 15, 2026

Added

• Customers can now ask for their exact prepaid balance from inside Hermes Startup (read-only balance action; public POST /v1/activate + GET /v1/wallet routes live on api.hermesstartup.com). Fails closed with no number when the control plane is unreachable.
• Homepage: "Browser included" — your agent can open a real browser and do the clicking for you (browser driver now provisions itself at install).

v0.1.13 August 14, 2026

Added

• Premium "Aurora Glass" pass: five-layer gold/amber/violet aurora field behind the hero (static, one-shot fade-in), frosted-glass activation card with backdrop blur, warm translucency and a bright top catch-light, gold hairlines on the dark section seams, and a restrained 3D tilt with a cursor-tracking gold glare on provider tiles (pointer-only, transition spring, no animation loops). All layers respect reduced-motion and reduced-transparency preferences; the original solid card remains the fallback without backdrop-filter support.

v0.1.12 August 13, 2026

Updated

• Backend upgraded to Hermes Agent 04d822211 — 407 upstream commits (193 fixes, 76 features) since v2026.8.3: plugin packs and manifest v2, gateway reliability fixes, cron hardening, desktop session resilience, and new plugin/desktop suggestion surfaces.

v0.1.11 August 12, 2026

Added

• Background depth layer: whisper of top-light on the page canvas, a warm gold-tinted shadow under the continuation panel, and a bookend glow behind the final activation section (all derived from the existing palette, flat black retained as the base).

v0.1.10 August 12, 2026

Added

• Premium motion layer expansion: word-by-word hero headline entrance, staggered scroll cascades for the 10-question list, continuation flow, capability chips and provider tiles, a slow synchronized count-up for the "1,000+ API tools" and "20+ providers" equation (both numbers finish at the same moment), a frosted sticky header, arrow micro-nudges on buttons and links, provider logo scale on tile hover, a gold ambient glow behind the hero, fine film grain over the page, and a cursor spotlight on the activation card. All one-shot, transform/opacity only, disabled under reduced motion, and content fully visible without JavaScript.

v0.1.9 August 11, 2026

Added

• Premium motion layer on the homepage: staggered hero fade-up entrance, a one-shot gold sheen on the copy button after a real successful copy, a factual "first step" check that pops in after copying, and subtle hover lifts on provider and category cards (transform/opacity only, safe under reduced motion, content fully visible without JavaScript).
• Provider grid collapses to a single column below 540px so provider names stay readable on mobile.

Removed

• The Evidence surface: header and footer links, the /outcomes.html page, and its references in llms.txt, llms-full.txt, for-agents.md, index.json, openapi.yaml, and sitemap.xml. The truthful outcome-claim boundary remains: no customer results, revenue, or testimonial claims without explicit permission and verifiable evidence.
• Below-the-fold scroll-reveal animations (IntersectionObserver-driven data-reveal behavior); sections are now always present as the user scrolls.

v0.1.8 August 11, 2026

Added

• A public Evidence page at /outcomes.html, linked from the site header and footer, with references from the machine-readable surfaces (llms.txt and others). Removed again in 0.1.9.

v0.1.7 August 4, 2026

Added

• The US$10 paid continuation is now live: checkout is via Stripe with a one-time prepaid balance, no subscription, and no automatic top-up; provider calls and external actions still require explicit approval. Public surfaces (llms.txt, llms-full.txt, for-agents.md, index.json, plugin metadata, and the in-workflow continuation preview) now state this consistently.
• Email authentication hardening: SPF now authorizes Google's sending servers (Gmail send-as) alongside Cloudflare Email Routing, and a DMARC monitoring record (p=none) with aggregate reports to hello@hermesstartup.com is published.
• Strict-Transport-Security (HSTS) enabled on the site, and a /.well-known/security.txt contact file published.
• Refreshed the 1200x630 social-sharing card: large readable headline, one supporting line, logo, and site URL (removed the tiny faux-interface box).

Changed

• Premium dark canvas across the homepage: subtle dot texture, ambient glows, hairline dividers, elevated paper card, and left-aligned footer links.

v0.1.6 August 4, 2026

Changed

• The Hermes Agent engine behind Hermes Startup was upgraded to v0.20.0. The guided path now runs on a faster, more reliable agent: conversational voice, grounded research with checkable citations, and much longer autonomous missions that recover from their own mistakes.
• Smarter context compression keeps long, multi-day journeys coherent.

v0.1.5 August 3, 2026

Fixed

• The public changelog page now renders code spans inside release notes as readable code instead of literal backticks.

v0.1.4 August 3, 2026

Added

• A public changelog page at /changelog.html, generated from this file, dated, and updated automatically on every release.

v0.1.3 August 2, 2026

Fixed

• Social-proof wording now always says "Hermes Agent users have copied this prompt" (never bare "Hermes"), with correct singular/plural forms, on the homepage, the public skill, and the runtime copy.
• Polling cost tuning: one check per minute before the 100-visitor gate, 5-second live refresh after, and no polling while the tab is hidden.

v0.1.2 August 2, 2026

Added

• Live social proof on the homepage: once 100 unique visitors are reached, the site shows "X Hermes Agent users have copied this prompt" with a real-time count and an arrow pointing at the copy button.
• Anonymous copy-event telemetry (Cloudflare Worker + Durable Object + D1): aggregate counts only, no prompt, email, clipboard, IP, or identity data is ever stored or returned.

v0.1.1 August 2, 2026

Changed

• Free tier now delivers 3 of 10 potential ideas, each explained in detail (why it fits you, how Hermes Startup would execute it, its potential, and a timeframe). These are a sample, not the top three.
• Paid continuation reframed: for US$10, Hermes Startup funds the API calls needed to start building your business for you, adds the remaining 7 ideas, and grades and ranks all 10 with the reason why.
• Billing model clarified: one-time prepaid balance, no subscriptions, no automatic top-up.
• Provider network positioning: 1,000+ API tools from 20+ providers, chosen and run for you from one balance.

v0.1.0 August 2, 2026

Added

• Initial public release of the reviewed Hermes Startup site at hermesstartup.com with a copyable activation prompt.
• Installable Hermes Startup skill for Hermes Agent (free, private 10-question onboarding that returns your unique ideas).
• One-paste Hermes activation and a local audit command.
• Portable public skill bundles (site-served and open-core GitHub mirror) that stay byte-identical.
• Legal pages (privacy, terms), agent resources (llms.txt, for-agents.md, sitemap), and social metadata.