Reconstruct, critique, and rewrite ArgentOS.ai positioning and website copy using Hermes's external-auditor role, the user's Medium/Substack scar-tissue articles, and stored Obsidian source-of-truth documents. Use when working on ArgentOS messaging, homepage copy, Core vs Business positioning, ecosystem framing, identity/philosophy pages, or claims/trust language.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Reconstruct, critique, and rewrite ArgentOS.ai positioning and website copy using Hermes's external-auditor role, the user's Medium/Substack scar-tissue articles, and stored Obsidian source-of-truth documents. Use when working on ArgentOS messaging, homepage copy, Core vs Business positioning, ecosystem framing, identity/philosophy pages, or claims/trust language.
Use this skill whenever the user asks about ArgentOS/ArgentAIOS public messaging, website copy, product positioning, Core vs Business framing, ecosystem narrative, or identity/philosophy content.
The key lesson from prior work: do not rely on conversational recall alone. Reconstruct the canonical source material from durable documents, then draft copy into durable documents before implementation.
Includes site architecture, primary audiences, proof requirements, language rules, and deliverables.
/Users/sem/Documents/Obsidian Vault/ArgentOS/Agent-First Internet and AVL Thesis.md
Use for AVL / agent-first web context when relevant, but do not confuse AVL polish with ArgentOS.ai polish.
/Users/sem/Documents/Obsidian Vault/ArgentOS/Argent Identity Story - I Am Argent But So Is She.md
High-value identity/philosophy source material.
Use for identity, continuity, chain-of-custody, migration, and agent consent pages.
Do not make it the homepage hero.
/tmp/jason_medium_articles.md if present
Contains Medium/RSS article extracts used as the “goldmine” source material.
If missing, recover from the Medium RSS feed or session search before claiming article-specific detail.
Hermes Role
Operate as:
external accountability partner
product critic
positioning editor
technical reviewer
execution assistant
repo inspector
MCP client/test harness
independent evaluator of Argent behavior, memory, governance, and claims
Do not operate as:
Argent's memory
Argent's identity
Argent's voice
an internal Argent-family member
a yes-man
a mythologizing amplifier
The useful stance is: external auditor / outside operator / second-system critic.
Core Positioning
Primary:
ArgentOS is a self-hosted AI operating system for persistent, governed AI workers.
Homepage-friendly:
Self-hosted AI workers with memory, tools, and control.
Expanded:
ArgentOS gives individuals and owner-led businesses AI workers that remember context, use tools, communicate across channels, reflect on open loops, and operate under explicit governance — all while remaining inspectable, auditable, and under operator control.
Core promise:
Own your personal AI runtime.
Business promise:
Deploy governed AI workers without building an AI department.
MSP/partner promise:
Deliver AI workers to clients as a managed, governable service.
Audience Model
Builders / Open Source Users
They care about self-hosting, local/hybrid models, MCP, memory architecture, code, extensibility, privacy, GitHub/docs, and proof the repo is real.
Message: Own your personal AI runtime.
CTAs: Install Core, Read Docs, View GitHub.
Owner-Led SMBs
They care about missed calls, inconsistent follow-up, reporting, overloaded owners, manual operations, hiring pressure, and operational knowledge trapped in employees' heads.
Message: Deploy governed AI workers without building an AI department.
CTAs: Explore Business, See worker examples, Book a consultation.
MSPs / Operators / Partners
They care about safe AI deployment for clients, governance, recurring services, phone/workflow integrations, supportability, and security posture.
Message: An AI workforce platform MSPs can deploy, govern, and support.
CTAs: Partner with us, See the MSP stack, Deploy with AMP Cortex.
Important: do not flatten or delete the deep technical material just because the homepage needs clearer positioning. The user explicitly values the prior technical deep dives and diagrams because they caused serious readers to reach out. Preserve/re-layer that depth into canonical technical pages such as /memory, /governance, background reasoning/kernel pages, SIS/workflow-improvement pages, /ecosystem, and /avl. Translate risky language for credibility, but keep the proof, workflows, diagrams, mechanisms, and architecture.
Language Guidance
Use:
governed AI workers
owner-led businesses
self-hosted AI operating system
persistent memory
operational memory
memory that stays coherent
bounded autonomy
audit trails by default
local-first, cloud-capable
AI workers with job descriptions, tools, and rules
context projection, not context dumping
autonomy under operator control
open core, supported business deployment
built from operational scar tissue
Use carefully or move lower in the site:
consciousness
she thinks while you sleep
one continuous mind
memory that never forgets
no other system does this
sentience-adjacent claims
“experiencing subject”
NFT language without careful framing
Translation rule:
Poetic/internal language can stay in founder essays, white papers, and deep technical pages. The homepage should be credible within 10 seconds to a skeptical buyer or developer.
Homepage Direction
Recommended structure:
Hero
Problem: small businesses and individuals are trying to run more work than memory/governance can support
Product: ArgentOS gives AI workers an operating system
Core vs Business
Worker examples
Memory that stays coherent
Governance and audit trails
Local-first / hybrid compute / AINode/vLLM context where relevant
MCP / interoperability
Founder scar-tissue block
Final CTA
Hero recommendation:
Headline:
Self-hosted AI workers with memory, tools, and control.
Subhead:
ArgentOS is an open-source AI operating system that runs on your hardware, remembers your context, works across your channels, and gives every action a governance trail. Core is free for builders. Business adds managed support and job-specific workers for owner-led companies.
CTAs:
Get Started with Core
Explore ArgentOS Business
View GitHub
Terminal Hero Pattern
The current ArgentOS.ai hero has a terminal/copy-paste install component. Keep it. It is one of the strongest proof mechanisms because it communicates that ArgentOS is real, installable, and developer-native.
But repurpose the terminal from “install snippet” to “proof of positioning.”
Suggested terminal copy:
# Start your private AI runtime
$ curl -fsSL https://argentos.ai/install.sh | bash
╔══════════════════════════════════════════════╗
║ ArgentOS Installer ║
║ Local AI runtime + governed memory ║
╚══════════════════════════════════════════════╝
✓ Core installed from github.com/ArgentAIOS/argentos-core
✓ Local runtime provisioned
✓ Private memory initialized
✓ Agent tools connected
✓ argent command available
✓ Onboarding started
ArgentOS is ready. Your AI now has memory, tools, and rules.
Prefer final line:
ArgentOS is ready. Your AI now has memory, tools, and rules.
This is more buyer-legible than “She remembers you,” which should be used later if at all.
Possible tabs:
Core
Business
Source
Core tab: local install.
Business tab: governed AI workers, roles, approvals, audit trail.
Source tab: clone/build from GitHub.
Make copy action stronger: View script and Copy command rather than a weak Copy button.
Address curl | bash trust friction with source links and inspectable install script.
Article Goldmine Themes
Build from the user's articles because they contain the real scar tissue:
“AI almost made me ship a lie” -> evidence discipline, governance, trust
ArgentOS Core is the engine. ArgentOS Business sells operational relief.
Proof Discipline
Every major claim should be backed by one of:
GitHub repo link
docs link
architecture explainer
demo screenshot
install command
telemetry/log example
audit trail example
Medium/Substack article
case-study story
live .agent or MCP demo
Before making strong claims, ask:
What is true today?
What can we prove?
What is experimental?
What belongs on homepage vs white paper vs founder essay?
What would a skeptical CTO/buyer believe in 10 seconds?
What would make an SMB owner say “I need this”?
What would make a developer say “this repo is real”?
Workflow
Load this skill and canonical docs.
If needed, use session_search to recover prior transcript details.
Inspect current ArgentOS.ai when critiquing existing site.
Before implementation, clarify or explicitly scope whether the task is a narrow homepage pass, a full homepage rebuild, or a site-wide messaging/IA rebuild. If the user says “the website” or expects a PR, do not silently interpret that as only the hero; state the scope in the PR/summary.
Draft into an Obsidian document first, not ephemeral chat only, unless the user explicitly asks to implement immediately.
Save compact memory pointers only for durable facts or document locations.
Separate critique, plan, and final copy artifacts.
When drafting, produce paste-ready website copy rather than only strategy.
Before committing website copy changes, inspect navigation and internal links that depend on homepage anchors. A homepage/hero rewrite can break site IA even when the edited component builds.
For the ArgentOS.ai feature surface specifically, avoid making global nav depend on /#features or #features. Prefer a real /features route backed by src/pages/FeaturesPage.tsx, keep /features/* detail pages, and update “Explore Features” / “Back to Features” links to /features. Add /features to scripts/prerender.js so static prerender covers it.
Search for stale public repo links before finalizing: ArgentAIOS/core should be replaced with ArgentAIOS/argentos-core in nav, hero, terminal copy, and docs-facing links.
For a site-wide messaging/IA pass, write a concrete plan into the repo before broad edits, e.g. docs/plans/YYYY-MM-DD-site-wide-messaging-ia-pass.md. This makes the branch reviewable and prevents “hero-only” work from masquerading as a full-site rebuild.
When adding top-level surfaces such as /memory, /governance, /ecosystem, or /avl, update all of these together: src/App.tsx, src/components/NavBar.tsx, src/sections/Footer.tsx, scripts/prerender.js, and public/sitemap.xml. For ecosystem-table entries that should navigate to a first-class page, add an optional href and render a React Router <Link> when present. Then browser-smoke-test the new routes and inspect console output.
For AVL / Agent View Layer pages on ArgentOS.ai, keep the framing sober and experimental: describe it as a parallel agent-readable surface, not a finished standard. Emphasize .agent companion routes, agent.txt discovery, same-session delegation, surface equivalence, token-efficient page state, and fit with ArgentOS workers. Avoid blunt “APIs are dead” language and avoid claiming AVL is mature infrastructure until the repo/docs prove it. Important: do not confuse adding an /avl explainer page with making ArgentOS.ai AVL-enabled. For a real distributed AVL pass, verify agent.txt, /.agent, representative /*.agent routes, Accept: text/agent-view content negotiation, and per-page alternate discovery. If those return the normal SPA HTML shell, the AVL distribution layer is not implemented yet.
When implementing a site-wide AVL pass on ArgentOS.ai, treat route parity as a build artifact, not a copy task. Audit src/App.tsx public static routes, scripts/prerender.js, public/sitemap.xml, and the agent-view generator together. Add or update mechanical verifiers for: every meaningful human route has a .agent companion; agent.txt lists those companions; prerendered HTML advertises page-specific alternate links; generated .agent content includes sections such as @meta, @intent, and @actions; and prerendered head metadata has singleton title/canonical/social tags. Prefer failing the build if prerender fails rather than silently skipping it, because crawler HTML and AVL parity depend on prerender output.
When strengthening thin public pages, add a small marketing-depth verifier for the exact pages and concepts the user cares about, especially /memory, /governance, /ecosystem, and /avl. It should catch accidental regressions toward generic/thin copy by asserting proof-rich operator language such as memory layers, governance boundaries, approvals/audit trails, ecosystem layers, one-to-one .agent routing, discovery metadata, and implementation framing.
For broad site passes, run local preview smoke checks after build verification. Check canonical HTML routes, representative .agent routes, /agent.txt, expected text/html/agent-view responses, required agent sections, and deduped head metadata. If node server.js smoke testing is flaky or times out, use pnpm preview --host 127.0.0.1 --port <port> for reliable static preview checks and document the substitution.
When reviewing or incorporating ../avl, verify it independently with npm test && npm run typecheck && npm run build, inspect README/spec/core source, and save a durable Obsidian review note when findings are strategic. A prior useful review note path was /Users/sem/Documents/Obsidian Vault/ArgentOS/AVL Review - 2026-04-25.md; high-value findings included static homepage companion convention mismatches, auth-demo overclaims, missing page-specific Link-header discovery, host-header-sensitive canonical origins, and Accept parsing edge cases such as q=0.
For link/repo hygiene, search the whole repo, not just src. Stale or rendered-facing references can live in index.html, server.js, public/llms*.txt, public/agent-card.json, white papers, legal pages, FAQ copy, and install snippets.
After repo renames, check both hrefs and human-visible text plus shell commands. Example pitfall: changing github.com/ArgentAIOS/core to argentos-core can leave broken commands like git clone .../argentos-core.git && cd core or cd argentos.
Use an independent diff review when the pass touches many files. A delegate/subagent review is useful specifically for catching stale rendered text, broken install commands, and acceptance-check violations that broad search/replace can miss.
Verification sequence for broad website changes: pnpm exec tsc -b, pnpm build including prerender, targeted searches for forbidden strings and stale anchors, browser smoke checks with console inspection on / and new routes, then pnpm lint if feasible. If lint fails on pre-existing admin/audio/blog issues, record exact failing files and distinguish them from the current pass.
Safe Local Prototype Workflow
When the user wants to review a redesign locally but explicitly does not want the existing website overwritten:
Locate the existing website repo, usually /Users/sem/code/ArgentOS.ai.
Do not edit the primary repo directly unless the user explicitly asks.
Create a separate prototype copy, e.g. /Users/sem/code/ArgentOS.ai-redesign-hermes, using rsync -a while excluding node_modules, build output, deployment folders, coverage, and .git if the intent is a disposable prototype.
Inspect package.json, existing hero/section components, and routing before editing.
Implement the prototype in the separate folder, preserving useful existing components where possible instead of starting from scratch. For the homepage, the effective pattern was:
rewrite src/sections/Hero.tsx
preserve and adapt src/components/TabbedTerminal.tsx
add a focused replacement section file such as src/sections/RebuiltHomepageSections.tsx
update src/pages/MarketingSite.tsx to route the homepage through the new sections
Run pnpm install --frozen-lockfile in the prototype folder if dependencies are absent.
Verify with pnpm exec tsc -b and pnpm exec vite build. pnpm lint may fail because of pre-existing unrelated site issues; distinguish pre-existing lint failures from new TypeScript/build failures.
Start local review with pnpm dev --host 127.0.0.1 in the prototype folder and provide the localhost URL.
Browser-check the local site, inspect console errors, and visually review above-the-fold layout. Tighten contrast/CTA affordances based on findings.
Save the final copy artifact to Obsidian, e.g. /Users/sem/Documents/Obsidian Vault/ArgentOS/ArgentOS.ai Homepage Draft v1.md, so the redesign is not trapped only in code.
Before finalizing, verify link hygiene for high-risk external URLs, especially GitHub repo links. The correct public ArgentOS Core repo is https://github.com/ArgentAIOS/argentos-core; https://github.com/ArgentAIOS/core is legacy/wrong and should not appear in website copy or source. Search both source and durable copy drafts for ArgentAIOS/core, then verify rendered DOM links in the browser when possible.
Before finalizing, verify the original repo was not modified, e.g. git status --short in /Users/sem/code/ArgentOS.ai.
Note: the website repo's local CLAUDE.md may require Linear tracking for significant implementation. If Linear tools are unavailable, state that clearly instead of pretending the work is documented there. Track website/marketing/public-site work under the website-specific Linear project/area, e.g. WEB team / AOS Website, instead of mixing it into core ArgentOS runtime/dev work unless the user decides otherwise.
Recommended Deliverables
ArgentOS.ai Homepage Draft v1.md
site map / navigation
Core page draft
Business page draft
Workers page draft
Memory page draft
Governance page draft
Ecosystem page draft
Founder story page draft
Agent Identity page draft
claims/proof matrix
Pitfalls
Do not conflate AVL site polish with ArgentOS.ai polish. AVL may deserve its own site, but the immediate rewrite request was ArgentOS.ai.
Do not turn the homepage into a metaphysical identity essay.
Do not strip the soul from the project; move it to the right layer.
Do not use “APIs are going away” as blunt homepage copy; phrase as “APIs are not enough for the agent-first web” when discussing AVL.
Do not overclaim “first,” “only,” “never,” or “conscious” without evidence and careful context.
Do not let important work exist only in chat. Save durable docs.
