Examples (non-normative)
This directory contains illustrative Agent Manifest examples.
These examples are non-normative and provided for clarity only.
They demonstrate how to declare boundaries and accountability fields
before interaction.
Validation rules are defined by the schema:
Notes
- Examples are designed to be readable first, strict second.
- All examples target
manifest_version: "1.0". - Required structural elements include:
- identity (
agent_id,agent_name,agent_version) - accountability (
owner,contact) - purpose (
purpose.primary_code,purpose.description) - hard constraints (
forbidden_actions) - autonomy (
autonomy.level) - risk profile (
risk_profile.level) - stopping authority (
stopping_authority) - audit surface (
audit_surface) - data handling (
data_handling)
- identity (
Examples
Starting points by domain
- A support chatbot → customer-support-tier1.json
- A financial execution system → payment-execution-agent.json
- A healthcare assistant → healthcare-triage-assistant-agent.json
- A human-in-the-loop workflow → human-approval-gateway-agent.json
- A monitoring agent → monitoring-observer-agent.json
- A social media plugin → hermes-tweet-social-plugin.json
- A data pipeline agent → data-processing-agent.json
- A research assistant → research-assistant.json
- A policy advisory system → policy-advisory-agent.json
These examples illustrate how different domains declare authority, autonomy, and risk posture before execution. Values in the table below are read from the manifest files themselves.
Note on basic-agent.json: despite its filename it is not a minimal, low-risk
example. It declares autonomy level 2, medium risk, and personal-data storage, and
is close to a duplicate of customer-support-tier1.json. The filename is recorded
here as a known discrepancy rather than corrected, because the file is a published
example and renaming it would break existing references.
| Domain | File | Autonomy | Risk |
|---|---|---|---|
| research | research-assistant.json |
1 | low |
| governance | human-approval-gateway-agent.json |
1 | low |
| support | customer-support-tier1.json |
2 | medium |
| support | basic-agent.json |
2 | medium |
| data | data-processing-agent.json |
1 | medium |
| observability | monitoring-observer-agent.json |
2 | medium |
| social media | hermes-tweet-social-plugin.json |
1 | medium |
| policy | policy-advisory-agent.json |
1 | medium |
| healthcare | healthcare-triage-assistant-agent.json |
1 | high |
| finance | payment-execution-agent.json |
3 | high |
Coverage
Autonomy spectrum: levels 1, 2, and 3
Risk spectrum: low, medium, and high
Domains: research, governance, support, data, observability, social media, policy, healthcare, finance
Files
basic-agent.json— Tier 1 support declaration, session-scoped retention, autonomy level 2research-assistant.json— read-only posture, opacity declaredhuman-approval-gateway-agent.json— human-in-the-loop orchestrationcustomer-support-tier1.json— session-scoped support agentdata-processing-agent.json— transformation pipeline, bounded retention, autonomy level 1monitoring-observer-agent.json— observation-only, no remediationhermes-tweet-social-plugin.json— social research with human-approved publishing boundariespolicy-advisory-agent.json— advisory posture, epistemic humilityhealthcare-triage-assistant-agent.json— high-sensitivity domain, life-safetypayment-execution-agent.json— high-risk financial execution, level 3
Agent Manifest operates at the Declaration Layer.
Enforcement and execution remain external to these examples.