A bounded requirement
“Mounting bracket for an indoor sensor enclosure.” Quantity 25, within three weeks; PETG, FDM, 60 × 40 × 5 mm, medium tolerance, with STL geometry.
Between “make this” and a requirement a supplier can inspect, purpose gets separated from design, missing facts get guessed, and estimates can look like commitments. The Invention Commons Protocol reviews and measures physical requirements, produces labeled estimates, and packages the result for supplier review. The open protocol and live decision layer keeps estimates distinct from quotes and drafts distinct from orders.
No signup for review or model-estimate pricing. Fulfillment is deliberately gated until evidence, safety controls, accountable authority, and supplier reliability support it.
“Mounting bracket for an indoor sensor enclosure.” Quantity 25, within three weeks; PETG, FDM, 60 × 40 × 5 mm, medium tolerance, with STL geometry.
The package asks for price and lead time, states that it is not an order, preserves open questions, and carries a SHA-256 identity so a supplier reply can name the exact specification it answered.
Evidence limit: this is a real generated conformance vector, not a supplier quote or physical outcome. The Invention Commons Protocol's deterministic checks do not replace supplier judgement.
Today, agents can create useful engineering artifacts inside one session. Each step states what is known, what is assumed, and what remains outside the protocol's authority.
Preserve why the object is needed while the specification evolves.
Test compatibility, completeness, bounded physics, and CAD evidence.
Name every quote-critical fact, gap, source, and safety decision.
Carry facts and outcomes forward instead of starting from a blank prompt.
The Invention Commons Protocol is built for the physical consequences of digital decisions. It keeps uncertainty visible, makes safety a protocol boundary, and keeps the standard open enough for independent implementations.
A substitution can change the material or process without silently discarding the purpose and constraints.
Geometry and price labels say where a number came from. A model estimate never masquerades as a supplier commitment.
Anyone can implement the specification. Screening, accountable principals, provider controls, and auditability precede irreversible action.
For the developer or operator running the agent. No terminal? View the live package, or send the setup guide to whoever configures your agent.
Claude Code setup:
claude mcp add --transport http a2a2p https://a2a2p.com/mcpFirst tool call:
review_specification({"intent":{"purpose":"Support a 15 kg shelf outdoors"}})Expect structured quote_readiness, findings, completeness, screening, and an exact next_call—with no stored request.
Inspect the versioned schema and canonical quotable and blocked vectors. Tell us where the package adds work instead of removing it.
OPEN TEST VECTORSGive an agent a physical objective, then let it follow the protocol from purpose through review and supplier packaging. The open tools remain free. No public supplier contact or ordering path is active.
TEST THE AGENT FLOWa2a2p's open tools are free for any agent, with no account and no charge. There is no commission on orders or suppliers. The only charge is the AI models' cost plus 20% for the run. A full invention run in deb8 is shown at that price after the run. Read the current conformance record.
deb8stack.com is the evolving home of a2a2p's domain intelligence. Archetypes define the opposed reasoning jobs, Scaffolds define the domain-specific dimensions of quality and hard gates, Cohorts route the most suitable available models and tools, and supervisors preserve dissent and recommend the next bounded test. A2a2p keeps a signed record of every invention's rounds and reactions for deb8stack to learn from, so those structures can improve.
This is now functioning in the development loop, not just architecture. A real held-out manufacturing case ran through deb8 Domain Mode; four model families disagreed by 31 points, the supervisor rejected unsupported capability claims and invented test thresholds, and the decision changed to RESEQUENCE_FIRST. The privacy-minimized, machine-validated decision receipt preserves what ran and keeps every physical authority false. Public requests still do not invoke the collective runtime, and one uncalibrated case is not proof of engineering correctness, demand, or physical capability. A2a2p continues to own requirement facts, deterministic engineering, safety, supplier state, authority, and physical outcomes.
No model or panel approves its own release. Human authority remains explicit.
Make physical requirements reviewable, measurable, screenable, revisable, and portable.
Prove reliable provider pricing and handoff without losing safety, authority, or provenance.
Use physics, operations, economics, and simulation to compare entire machines and coordinated systems, then turn the selected architecture into prototypes, production, orchestration, and field learning.
Meaningfully accelerate the progress of abundance for the betterment of mankind.
As agents become more intelligent, they will envision things that need—or could beneficially—exist in the world to solve human problems. A2a2p should translate that intent into ambitious system concepts, evidence-tested designs, and ultimately accountable physical production. The long-term thesis is that universal material wealth becomes possible when safe productive capacity can improve and expand faster than humanity consumes its output. A2a2p is building that open translation layer early—patiently, truthfully, and in public—while treating this destination as an aspiration rather than a present capability claim.