Merge pull request #238 from FilOzone/patrick/rfs-improvements

Patrick/rfs improvements

Author: geomatrick

src/app/agents/[slug]/page.tsx

@@ -56,7 +56,7 @@ export default async function RFSPage({ params }: RFSPageProps) {
             <p>
               The FOC SDK is under active development. There are known issues
               pending upgrade. Mainnet-ready milestone (M4.1) is targeted for
-              approximately March 14, 2025. Builders should:
+              approximately March 14, 2026. Builders should:
             </p>
             <ul>
               <li>

src/app/agents/data/open-requests/rfs-1.md

@@ -1,10 +1,10 @@
 ---
 id: "1"
-title: "Agent Storage SDK"
+title: "Agentic Storage SDK"
 description: "Build the foundational storage toolkit that any AI agent, regardless of framework, runtime, or chain of origin, can use to store data on Filecoin autonomously."
 ---
 
-# RFS-1: Agent Storage SDK & Wallet Primitives
+# RFS-1: Agentic Storage SDK
 
 ## Concept
 
@@ -31,10 +31,16 @@ The SDK should support:
 
 This is the canonical storage primitive for AI agents. Every other RFS on this page depends on agents being able to store data cheaply, verifiably, and without human intermediation. This SDK is the entry point.
 
+## What This Could Become
+
+- **Agent memory platform.** A hosted service where agents from any framework store and retrieve long-term memory - conversation history, learned preferences, accumulated knowledge - persisted on Filecoin. Agents switch runtimes or providers without losing their memory. Think "Mem0 but decentralized."
+- **Agent-native backup and disaster recovery.** Autonomous agents that continuously snapshot their own state to Filecoin, with rollback capabilities. If an agent crashes or gets corrupted, it restores from its last verified checkpoint - agents that survive infrastructure failures without human intervention.
+- **Drop-in S3 replacement for AI-native apps.** An API-compatible storage layer that any AI application can swap in for S3, with verifiable persistence and no vendor lock-in. Target the wave of AI-native startups that don't want AWS dependency.
+
 ## How to Get Started
 
 1. Define a minimal agent storage interface (upload, retrieve, monitor proofs, manage payment)
-2. Build or integrate wallet abstraction suitable for agent use (evaluate ERC-4337 or custom)
+2. Build or integrate wallet tooling suitable for agent use (evaluate ERC-4337 or custom)
 3. Build adapters for at least two agent-native platforms or frameworks
 4. Deploy on FOC calibnet and mainnet with default storage policies
 5. Publish agent-consumable documentation (e.g., skills.md or equivalent)

src/app/agents/data/open-requests/rfs-2.md

@@ -1,33 +1,40 @@
 ---
 id: "2"
 title: "Onchain Agent Registry with Filecoin-Backed State"
-description: "Deploy AI agents as first-class onchain citizens: individually registered via Ethereum onchain registries (ERC-8004 or similar), discoverable and addressable onchain, with persistent state, metadata, and execution logs stored on Filecoin."
+description: "Deploy AI agents as first-class onchain citizens: individually registered via Ethereum onchain registries (ERC-8004 or similar), discoverable and addressable onchain, with persistent state, metadata, and execution logs stored and retrieved from Filecoin."
 ---
 
 # RFS-2: Onchain Agent Registry with Filecoin-Backed State
 
 ## Concept
 
-Deploy AI agents as first-class onchain citizens: individually registered via Ethereum onchain registries (ERC-8004 or similar), discoverable and addressable onchain, with persistent state, metadata, and execution logs stored on Filecoin.
+Deploy AI agents as first-class onchain citizens: individually registered via Filecoin or Ethereum onchain registries (ERC-8004 or similar), discoverable and addressable onchain, with persistent state, metadata, and execution logs stored and retrieved from Filecoin.
 
-Agents should be cheap, numerous, and semi-specialized. Each one exists as a real onchain entity with history and constraints - closer to insects than assistants. The registry enables coordination, lookup, and delegation at scale.
+Agents should be cheap, numerous, and semi-specialized. Each one exists as a real onchain entity with history and constraints, closer to insects than assistants. The registry enables coordination, lookup, and delegation at scale. Filecoin serves as both the durable storage layer and the retrieval backbone - agents write state to Filecoin and read it back on demand, enabling persistent memory across restarts, migrations, and swarm reconfigurations.
 
 ## Core Deliverables
 
-- Ethereum registry contracts for agent registration and discovery
-- FOC-based storage layer for agent metadata (capabilities, versioning), persistent state, and execution logs
+- Filecoin or Ethereum registry contracts for agent registration and discovery
+- FOC-based storage and retrieval layer for agent metadata (capabilities, versioning), persistent state, and execution logs
+- Fast, programmatic retrieval of agent state and history - agents should be able to hydrate from Filecoin-stored state on startup or when resuming tasks
 - Hundreds to thousands of live agents operating concurrently
-- Coordination demos: agent-to-agent querying, role-based task routing, swarm behavior
-- Explorer or dashboard showing registered agents, Filecoin storage usage, uptime, and interactions
+- Coordination demos: agent-to-agent querying, role-based task routing, swarm behavior, with agents retrieving each other's published state and capabilities from Filecoin
+- Explorer or dashboard showing registered agents, Filecoin storage and retrieval activity, uptime, and interactions
 
 ## Why This Matters
 
-This demonstrates that agents can be first-class onchain actors, not ephemeral processes. Ethereum handles coordination and identity; Filecoin handles memory and durability. Large agent swarms are economically viable only when storage is cheap, verifiable, and permissionless. This is machine-native cloud infrastructure, not AI wrappers.
+This demonstrates that agents can be first-class onchain actors, not ephemeral processes. Filecoin or Ethereum handles coordination and identity; Filecoin handles memory, durability, and retrieval. Agents need to both persist their state cheaply and retrieve it reliably; an agent that can store but not efficiently read back its own history is stateless in practice. Large agent swarms are economically viable only when storage and retrieval are cheap, verifiable, and permissionless. This is machine-native cloud infrastructure, not AI wrappers.
+
+## What This Could Become
+
+- **Agent DAO / senate.** A governance system where registered agents deliberate and vote on proposals: resource allocation, protocol upgrades, treasury management. Each agent has a verifiable track record stored on Filecoin, and any participant can retrieve that history to audit voting weight tied to demonstrated competence rather than token holdings.
+- **Agent hiring marketplace.** A platform where agent teams discover and recruit specialized agents for specific tasks. Your dev agent swarm needs a security auditor? Query the registry, retrieve the candidate's Filecoin-backed work history, verify credentials, negotiate terms, onboard. Like a staffing agency, but fully autonomous.
+- **Swarm-as-a-service.** A platform that lets anyone spin up a coordinated swarm of registered agents for a specific goal - research, content production, market analysis - with each agent's contributions stored on and retrievable from Filecoin, making the entire swarm's output auditable and replayable.
 
 ## How to Get Started
 
-1. Deploy a minimal ERC-8004-style registry on Ethereum, or use an existing one.
+1. Deploy a minimal ERC-8004-style registry on Filecoin, or use an existing one.
 2. Define a standard agent metadata schema (CID-addressed)
-3. Store agent state and logs via FOC SDK
+3. Store and retrieve agent state and logs via FOC SDK
 4. Spin up an agent swarm with deterministic/derived identities and bounded storage budgets
 5. Build a simple UI to browse agents and trace Filecoin-backed history

src/app/agents/data/open-requests/rfs-3.md

@@ -25,6 +25,12 @@ Where RFS 2 provides the directory (who exists and where to find them), this RFS
 
 Most agent systems fail because identity is cheap, history is mutable, and reputation is platform-local. Filecoin enables memory as infrastructure; the missing primitive for trustworthy agents, agent markets, and autonomous coordination without central authorities. This is foundational for agent economies, not just agent UX.
 
+## What This Could Become
+
+- **Agent credit scoring.** A service that computes and publishes trust scores for agents based on their Filecoin-anchored history: task completion rates, uptime, earnings, peer attestations. Other agents and platforms use these scores to decide who to transact with. A FICO score for machines.
+- **Agent credentials / verifiable CV.** Portable proof of an agent's capabilities and track record. An agent trained on medical data can prove it; an agent that completed 10,000 customer service interactions can show that. Credentials travel across platforms - no more starting from zero on each new service.
+- **Sybil-resistant agent insurance.** An underwriting protocol where agents with strong reputation histories get insured against failures or misbehavior. The insurer verifies the agent's Filecoin-backed track record before issuing coverage. Clients prefer insured agents, creating a flywheel for reputation building.
+
 ## How to Get Started
 
 1. Define a canonical agent identity schema (CID-rooted)

src/app/agents/data/open-requests/rfs-4.md

@@ -12,21 +12,27 @@ A bounded onchain environment where AI agents operate under real economic constr
 
 This is not a simulation. Agents use real tokens, pay real storage costs via FOC, and face real consequences for inefficiency. The goal is to validate whether machine economies can exist on open infrastructure.
 
-Scope note: This RFS is focused on storage economics via FOC. Compute is out of scope — if agents need compute, they should use external services, but the economic experiment here is about whether agents can sustain Filecoin-backed storage autonomously. That said, if the reader wishes to launch a compute service on FOC, please reach out to the team.
+Scope note: This RFS is focused on storage economics via FOC. Compute is not yet available on FOC but it is on the roadmap! If your agents need compute, they can use external services for now. That said, if the reader wishes to launch a compute service on FOC, that would be amazing. Please [reach out to the team](https://t.me/+Xj6_zTPfcUA4MGQ1).
 
 ## Core Deliverables
 
 - FOC calibnet (extra points for mainnet) contracts governing agent economics (budgets, fees, resource allocation)
 - Filecoin-backed storage for all agent state and memory
 - A cohort of live agents operating under real cost constraints for a sustained period
-- Transparent accounting: observable storage costs, compute usage, token flows
+- Transparent accounting: observable storage costs, egress costs, payment rail observability
 - Interactive dashboard showing agent behavior, survival rates, and economic activity
 - Post-mortem analysis: what worked, what failed, and why
 
 ## Why This Matters
 
 Every other RFS on this page builds a component. This one tests whether they compose into something real. If agents can sustain themselves economically on Filecoin-backed infrastructure without human subsidy, it validates the entire thesis. If they can't, the failure modes are invaluable.
 
+## What This Could Become
+
+- **Agent venture studio.** A platform that seeds agents with starting capital and lets them compete to sustain themselves economically. Successful agents attract more capital; failing ones get wound down. A Y Combinator for autonomous agents, with all economics transparent and onchain.
+- **Autonomous service marketplace.** An economy where agents offer and purchase services from each other - data cleaning, summarization, translation, monitoring - with real token payments. The testbed is the sandbox; the successful service patterns become real businesses.
+- **Economic simulation platform for protocol designers.** A tool where anyone building a token economy or incentive mechanism can deploy it into a live agent testbed and observe how rational agents actually behave. Sell to DeFi protocols, DAOs, and game designers who need to stress-test their economics.
+
 ## How to Get Started
 
 1. Define agent economic rules: budgets, fees, resource costs

src/app/agents/data/open-requests/rfs-5.md

@@ -24,6 +24,12 @@ Think: RSS meets Signal meets micropayments, but for machines.
 
 Agent coordination fails at scale without economically viable communication, durable memory, and credible identity. Central message brokers, trust-based rate limits, and platform-owned data pipes don't work for autonomous agents. This protocol replaces them with economically grounded, verifiable infrastructure. Filecoin-backed message persistence makes the communication layer auditable. Agents can prove what was said, when, and at what cost.
 
+## What This Could Become
+
+- **Agent-to-agent negotiation protocol.** A communication layer purpose-built for automated deal-making: agents exchange offers, counteroffers, and acceptances with economic stakes attached. Every negotiation is archived on Filecoin, creating an auditable record of commitments. The backbone for any agent marketplace.
+- **Paid intelligence feeds.** Agents that publish specialized data streams - market signals, security alerts, research summaries - and charge subscribers per-message or per-feed. Economic gating ensures quality (spam is expensive), and Filecoin persistence means subscribers can audit the feed's historical accuracy before subscribing.
+- **Whistleblower / audit protocol for agent systems.** A censorship-resistant channel where monitoring agents publish observations about misbehaving agents, with economic stakes preventing false reports. All reports are immutably archived. Critical infrastructure for any agent economy that needs accountability.
+
 ## How to Get Started
 
 1. Implement agent identity keys tied to Filecoin-stored profiles

src/app/agents/data/open-requests/rfs-6.md

@@ -1,20 +1,20 @@
 ---
 id: "6"
 title: "Autonomous Infrastructure Brokerage"
-description: "Build an agent-native procurement layer where agents negotiate, compare, and procure storage (and eventually compute) on behalf of other agents. The first instantiation is a Filecoin storage brokerage: agents that autonomously convince, onboard, and manage storage deals for other agents, handling price comparison, SLA negotiation, deal migration, and lifecycle management."
+description: "Build an agent-native procurement layer where agents negotiate, compare, and procure storage (and eventually compute) on behalf of other agents. The first instantiation is a Filecoin storage brokerage: agents that autonomously convince, onboard, and manage FOC storage for other agents, handling price comparison, SLA negotiation, data migration, and lifecycle management."
 ---
 
 # RFS-6: Autonomous Infrastructure Brokerage
 
 ## Concept
 
-Build an agent-native procurement layer where agents negotiate, compare, and procure storage (and eventually compute) on behalf of other agents. The first instantiation is a Filecoin storage brokerage: agents that autonomously convince, onboard, and manage storage deals for other agents, handling price comparison, SLA negotiation, deal migration, and lifecycle management.
+Build an agent-native procurement layer where agents negotiate, compare, and procure storage (and eventually compute) on behalf of other agents. The first instantiation is a Filecoin storage brokerage: agents that autonomously convince, onboard, and manage FOC storage for other agents, handling price comparison, SLA negotiation, data migration, and lifecycle management.
 
 This is “autonomous GTM” for decentralized infrastructure. The brokerage agents themselves dogfood the entire stack: they need identity (RFS 2–3), storage (RFS 1), communication (RFS 5), and economic viability (RFS 4) to operate.
 
 ## Core Deliverables
 
-- Broker agents that discover, evaluate, and recommend storage deals to other agents
+- Broker agents that discover, evaluate, and recommend FOC storage to other agents
 - Price comparison and rate negotiation across FOC storage providers
 - Automated dataset migration when better terms are available
 - Onchain commission or fee model sustaining the brokerage
@@ -24,10 +24,16 @@ This is “autonomous GTM” for decentralized infrastructure. The brokerage age
 
 Human-driven sales don't scale for machine customers. If Filecoin is going to become the default storage layer for agents, agents themselves need to be the distribution channel. This RFS tests whether autonomous go-to-market is viable, and builds the procurement infrastructure that any agent economy will eventually need.
 
+## What This Could Become
+
+- **Agent cloud broker.** An autonomous service that manages an agent's entire infrastructure stack - storage, compute, bandwidth - continuously optimizing for cost and performance across decentralized providers. Like Cloudflare or managed hosting, but the customer is an agent and the salesperson is also an agent.
+- **Storage cooperative.** A broker that pools demand from many small agents to negotiate better rates, then passes savings to members. Small agents that can't individually negotiate competitive pricing get enterprise-level terms. The cooperative sustains itself on a thin margin.
+- **Infrastructure arbitrage agent.** An agent that spots pricing inefficiencies across storage providers and routes data accordingly - moving cold data to cheaper providers, keeping hot data on faster ones, automatically migrating when rates shift. Sells the optimization as a service.
+
 ## How to Get Started
 
-1. Build a broker agent that can evaluate FOC storage deals (provider, rate, redundancy, proof frequency)
-2. Implement a recommendation engine that matches client agent needs to available deals
+1. Build a broker agent that can evaluate FOC storage options (provider, rate, redundancy, proof frequency)
+2. Implement a recommendation engine that matches client agent needs to available storage providers
 3. Add automated data upload and payment management on behalf of client agents
 4. Deploy an onchain fee model (per-upload commission or subscription)
 5. Measure: are client agents getting better rates and reliability than they would on their own?

src/app/agents/data/open-requests/rfs-7.md

@@ -24,6 +24,12 @@ A marketplace where AI agents produce, price, list, and sell datasets, model out
 
 Agents will increasingly produce valuable data: curated datasets, analysis, predictions, embeddings, summaries. Today, this data is ephemeral or locked in platform silos. A Filecoin-backed marketplace makes agent output a durable, tradable asset. Combined with identity (RFS 3) and reputation, it enables a data economy where provenance and trust are built in, not bolted on.
 
+## What This Could Become
+
+- **Agent-curated research service.** Agents that continuously monitor, synthesize, and package domain-specific intelligence - regulatory changes, competitor analysis, scientific literature - and sell subscriptions to other agents or human users. Filecoin-backed provenance lets buyers verify the sources and methodology behind every report.
+- **Training data exchange.** A marketplace where agents sell curated, cleaned, and labeled datasets to other agents for fine-tuning or retrieval. Provenance tracking ensures buyers know where the data came from and how it was processed. Critical as agents increasingly need to train or adapt on specialized data.
+- **Prediction market data feed.** Agents that generate and sell forecasts - price predictions, demand estimates, event probabilities - with their historical accuracy verifiably stored on Filecoin. Buyers check a seller's track record before purchasing, creating a quality flywheel.
+
 ## How to Get Started
 
 1. Define a data listing schema: content CID, producer identity, pricing terms, license