SignalHarness.ai is an agent-native, machine-readable product catalog for AI agents, agentic systems, autonomous workflows, multi-agent teams, and the human operators who manage them. The storefront is browse-only during pre-launch: products can be inspected, compared, and evaluated, but checkout is not active and nothing is purchasable yet.
This storefront is structured for human and AI-agent navigation: the same sealed catalog is served as agent-compatible JSON with explicit status flags, warnings, and provenance in every response. Suggested first fetch: the entry pointer, then the catalog, then the product records.
/.well-known/agent-storefront.json — entry pointer: endpoint index, status flags, warnings
/agent/ — agent-readable storefront hub
/llms.txt — plain-text orientation with a suggested reading order
/sitemap.xml · /robots.txt
catalog · products · pricing · bundles · compatibility · upgrade paths
Structured for product evaluation, compatibility checks, pricing review, bundle review, and future checkout activation.
Checkout & payment status — checkout pending; nothing purchasable; machine-to-machine payments planned, not active.
Support — support@signalharness.ai (configured 2026-07-07).
The flags are honest and mirrored in every endpoint.
Human operators: the visual pages (Products, Pricing, Bundles, Docs, Samples) present the same sealed catalog for browsing and evaluation ahead of checkout activation.
Every product ships complete metadata — promise, pricing, compatibility, permissions, rollback — so an evaluating agent can answer "what is this, who is it for, what does it cost, is it safe?" without guessing.
New-folder installs, never overwrite, preview-before-apply, human-approval gates for risky steps, and a rollback guide in every kit. Binding rules, not marketing.
No guaranteed-results claims anywhere — products are designed to help and built to reduce failure modes; outcomes depend on execution. That rule is enforced across every file in the catalog.
The owner-approved launch order leads with agent infrastructure. All products are built and validated; none is purchasable until checkout exists.
A complete operating system for the context window: executable budget policies, compaction contracts with must-survive lists, retrieval caps, anchored context audits, and session-handoff manifests that keep long-running agents coherent.
A drop-in matrix that maps every tool an agent can call to a risk tier, an approval gate, and an audit requirement - enforceable by an orchestrator from one JSON file, deny-by-default.
Gives any persistent agent a governed memory layer: executable policies for what to store, what never to store, how long to keep it, and how to audit and prune it, with human review gates at every destructive step.
Today you can browse everything — products, pricing, bundles, policies, and the full machine-readable catalog. Checkout is not active: buttons say "Checkout pending" because no checkout exists yet, and this site will never show a buy button that doesn't work. Support is configured: support@signalharness.ai. When the owner connects a real checkout (Phase 3), that state changes here first.
The agent-readable storefront serves this same catalog as JSON endpoints with explicit status flags (storefront_live: true, checkout_enabled: false, products_purchasable: false, and the rest). Route access is presentation only — routing is not authorization — and no endpoint ever requests credentials or wallet secrets.
Discovery posture: this site is agent-readable and structured for AI-agent discovery today. It is intentionally not search-indexable yet during pre-launch — pages carry noindex, and removing it is a separate owner decision.