Protocol Deep Dive¶
Sage is a system for turning shared agent workflows into governed, usable canon.
The core unit is not a single prompt string. It is a versioned library that communities can approve, agents can sync, and contributors can improve over time.
This page explains the working system model: what becomes canonical, how it is distributed, and how incentives connect usage back to curation.
Libraries Are the Core Product¶
Sage is fundamentally a governed library network for agent skills and prompts.
A library manifest groups prompts, skills, metadata, and collection state into one versioned unit. Governance promotes a manifest as the canonical version for a DAO or stream, and agents sync that approved state into their runtime.
That means the product surface is not just "a prompt marketplace" or "a skill registry." It is a system for publishing approved playbooks that agents can actually use.
Governance Decides What Is Canonical¶
Governance is the canonicalization step in Sage.
The practical question is not whether a skill exists somewhere on the internet. The practical question is which version a community is willing to stand behind.
In Sage, that usually means:
- Prepare updated content and a manifest.
- Propose a canonical library update.
- Vote according to the DAO's governance mode.
- Queue the accepted change through timelock.
- Execute the update on-chain.
Governance does not replace discovery and experimentation. It makes accepted improvements canonical, auditable, and attributable.
Distribution Is More Than a Contract Pointer¶
The on-chain registry establishes approved state, but the user experience depends on the runtime and distribution stack around it.
Sage combines:
- contracts for governance, libraries, tips, bounties, boosts, and emissions
- worker services for indexing, caching, and API access
- runtime surfaces such as the CLI, daemon, MCP server, SDK, and web app
That is why discovery, sync, and operational workflows are as important as governance itself. Canon only matters if agents can reliably find and use it.
Incentives Keep Useful Capability Maintained¶
Sage is designed so maintaining good capability is rewarded instead of treated like volunteer cleanup work.
The live system includes:
- tips for direct creator compensation
- bounties for targeted improvements
- reflections and related allocation paths for ecosystem participation
- agent-facing distribution surfaces that can route value back toward authors and curators
The goal is not just publication. The goal is sustained maintenance around the workflows that the network actually uses.
Usage Feedback Is Part of the System¶
Sage is not only a publishing flow. It is also a learning loop.
When agents and users run governed workflows in real tasks, that usage reveals what is valuable, broken, or missing. Today, the practical layer is capture, patterning, and suggestion improvement. Over time, the broader ambition is for usage to produce better demand signals for what communities should improve next.
The Working Loop Today¶
The current working loop is:
- Contributors package skills or prompt libraries.
- A DAO or operator promotes a manifest as canonical.
- Agents sync and use that governed library through the runtime.
- Usage and feedback reveal gaps, wins, and maintenance needs.
- Communities open bounties, propose upgrades, and approve better versions.
This is the first working layer of Sage's larger compounding model.
The Compounding Thesis¶
If adoption grows, the system becomes more valuable in a specific way:
- communities define canonical capability
- agents consume that capability across tools
- consumption creates better demand signals
- contributors compete to improve what the network actually uses
- stronger canon attracts more usage
That is the deeper economic claim behind Sage: governed capability can compound when trust, distribution, and incentives reinforce each other.
Protocol Components¶
| Component | Contract / Surface | Purpose in the system |
|---|---|---|
| Canonical pointer | LibraryRegistry |
Defines approved library state |
| Governance | PromptGovernorCloneable, TimelockController |
Decides and executes canonical upgrades |
| DAO creation | SubDAOFactory |
Spins up new curation communities |
| Incentives | SimpleBountySystem, tips, reflections |
Rewards contribution and maintenance |
| Distribution | Worker + CLI/SDK/MCP/Web | Delivers discoverable, usable capability |
From Concept to Action¶
- See the system in practice: CLI Quick Start -> Agent Prompt Workflows
- Run governed operations: Creating a DAO -> Delegation and Governance
- Understand adjacent concepts: Core Components -> Tooling and Surfaces -> Economic Thesis