Loading...
Not prompts. Not agents. Contracts you can trust, diff, test, and sign. The governance layer for the agentic era.
---
proto: 0.1
name: "Shopping Compare Helper"
capabilities:
mode: "browser_assist"
autonomy:
level: "assisted"
ask_before: ["submit_forms", "spend_money"]
permissions:
data_access:
payments: "deny"
passwords: "deny"
scopes:
allowed_domains: ["amazon.in", "flipkart.com"]
denied_actions: ["checkout", "place_order"]
---As agents gain the ability to run commands, browse the web, and access tools — risk and ambiguity scale faster than intelligence.
Agents act in black boxes. No visibility into what they do, access, or refuse.
Overbroad access to files, shell, network, and credentials with no standard contracts.
Model updates and tool changes silently shift behavior. No tests, no baselines.
No signing, no provenance, no way to verify who published an agent or what changed.
Three steps to verifiable agent behavior.
Write a proto.md with objectives, permissions, scopes, and safety rules — or use the guided builder.
Run ProtoLint to catch dangerous permission combos, missing safety gates, and contract contradictions.
Publish to the registry, embed in your repo, or export to any agent runtime.
From spec to signing — the full governance toolkit for agent contracts.
One file, one truth. YAML frontmatter for machines, markdown for humans.
Static analysis for agent contracts. Flags dangerous permission combos and missing safety gates.
Guided wizard that generates valid proto.md files with safe defaults. No spec knowledge needed.
Browse curated, linted, and verified agent recipes. Filter by risk, permissions, and task.
Behavioral change logs. See exactly what permissions, scopes, and refusals changed between versions.
Cryptographic signing for agent contracts. Verified publishers, version integrity, org approvals.
Define what your agent does, what it won't do, and prove it.