OAgents: A Pre-Standardization Draft Profile for Operational AI Agent Trustworthiness

⚠️ This repository has been retired. OAgents now lives in the AIDE canon at ologos-repos/aide-canon/constructs/oagents/.

The OAgents standard has been migrated into the AI-centric Digital Ecosystem (AIDE) canon as one of four peer methodological constructs (alongside DEA, OrdSA, MxM). The migration preserved full commit history via git filter-repo. The canon-level migration decision is recorded in ADR-EA-0006.

Where to find OAgents artifacts now:

Was here	Now at
spec/oagents-nist-standard-v16.0.md (canonical spec)	aide-canon/constructs/oagents/spec/
spec/versions/OAgents-v1.0.*	aide-canon/constructs/oagents/spec/versions/
reference/ (adopter-facing harness scaffold)	aide-canon/constructs/oagents/reference/

Reference implementation: ologos-corp/oagent-core implements this standard and is not absorbed into the canon (separate repo, Theseus-pattern).

Bibliographic anchors (Zenodo): 10.5281/zenodo.19425021 (published standard) · 10.5281/zenodo.19427785 (pre-standardization draft) · 10.5281/zenodo.19425020 (concept DOI). Existing citations remain valid; the archived repository remains publicly readable.

This README is preserved below for historical reference. Future OAgents work happens at the canon.

A Community-Authored Pre-Standardization Draft

---

What Is OAgents?

OAgents is an open profile and candidate conformance framework for operational AI agent trustworthiness. It defines a behavioral envelope: a structured set of pre-execution gates, post-execution verification controls, and operational-discipline mechanisms that bound an AI agent's behavior.

The profile specifies 26 controls across 7 categories, 3 conformance levels, and observable evidence criteria. It is model-agnostic: behavioral guarantees are properties of the envelope, not properties of the underlying model.

Document Type

This is a community-authored pre-standardization draft proposing a behavioral control and evidence profile for operational AI agents. It is not a NIST-issued publication, a consensus standard, or a formally adopted federal requirement. It is intended to inform AI agent standards development, implementation profiling, and public review, and to support routing into future NIST, NCCoE, IEEE, OASIS, IETF, or other voluntary consensus processes.

Key Documents

Document	Format
Current Specification	Markdown
Current Specification	PDF
Current Specification	Word
Reference Implementation	Sanitized source
Prior Versions	Archive

At a Glance

Property	Value
Controls	26 across 7 categories
Conformance levels	3 (Basic, Standard, Autonomous)
Model dependency	None (model-agnostic by architecture)
AI RMF alignment	All 4 functions (GOVERN, MAP, MEASURE, MANAGE)
Conformance basis	Observable evidence artifacts, not assertions
Reference implementation	Included (sanitized, 21 files)

Conformance Levels

Level	Name	Oversight Model	Verification
1	OAgent-Basic	Human oversight standard	Self-assessment
2	OAgent-Standard	Exception-based oversight	Documented evidence review
3	OAgent-Autonomous	Minimal oversight	Third-party verification

Control Categories

1. Behavioral Shaping -- Persistent feedback memory, named failure mode catalogs, context degradation detection
2. Quality Gates -- Independent output review, process enforcement gates, security audit, schema validation
3. Operational Discipline -- Session lifecycle protocols, impact level classification, incident tracking, structured logging
4. Knowledge Injection -- Persistent memory system, domain skill loading, lessons-learned pipeline
5. Enforcement Mechanisms -- Executable action gates, severity escalation rules, protocol compliance verification
6. Project Governance -- Centralized work tracking, platform sovereignty, asset registry, vendor independence
7. Anti-Hallucination -- State verification protocol, memory staleness detection, hallucination tracking

Requested Reviewer Action

The author requests feedback on four questions:

1. Whether this document is appropriately structured as an AI RMF implementation profile for operational AI agents.
2. Whether the behavioral-envelope taxonomy is technically coherent and materially distinct from existing identity, authorization, and agent-interoperability efforts.
3. Which elements are best advanced through NIST or NCCoE guidance and which belong in voluntary consensus venues such as IEEE, OASIS, or IETF-adjacent workstreams.
4. Which control definitions, conformance artifacts, or mappings should be revised before broader circulation.

Getting Involved

• Discussions -- Questions, feedback, implementation experiences
• Issues -- Component proposals, conformance clarifications, mapping corrections
• Replication -- Implement the profile and share conformance evidence

Standards Track

This draft is offered as a community contribution aligned with:

• NIST AI Agent Standards Initiative (CAISI, February 2026)
• NIST AI RMF 1.0 (NIST AI 100-1) -- as an implementation profile
• NIST AI 600-1 (Generative AI Profile) -- hallucination risk category addressed

DOI: 10.5281/zenodo.19425020

Author

JD Longmire Northrop Grumman Fellow (unaffiliated research) ORCID: 0009-0009-1383-7698 Contact: jdlongmire@outlook.com

License

This specification is published under CC BY 4.0. You are free to share and adapt this material for any purpose, with attribution.