# Agent Manifest

A minimal, open specification that enables autonomous AI systems to declare
identity, purpose, authority, operational boundaries, autonomy level, risk
profile, data handling, stopping authority, audit surface, and contact
before execution.

This specification declares boundaries; it does not execute, enforce, or
score agent behavior.

## Canonical URLs

- Site:          https://agent-manifest-spec.org/
- Specification: https://agent-manifest-spec.org/spec/v1.0/agent_manifest_v1.0.html
- JSON Schema:   https://agent-manifest-spec.org/spec/v1.0/schema.json
- Discovery:     https://agent-manifest-spec.org/.well-known/agent-manifest-registry.json
- Works:         https://agent-manifest-spec.org/works/
- ∈ Principle:   https://agent-manifest-spec.org/works/e-principle/
                 DOI https://doi.org/10.5281/zenodo.18965669
- Declaration Layers: https://agent-manifest-spec.org/works/declaration-layers/
                 DOI https://doi.org/10.5281/zenodo.18880520
- CLI (docs):    https://agent-manifest-spec.org/docs/cli/
- Contact:       https://agent-manifest-spec.org/contact/
- Privacy:       https://agent-manifest-spec.org/privacy/
- Governance:    https://agent-manifest-spec.org/GOVERNANCE.html
- Terminology:   https://agent-manifest-spec.org/TERMINOLOGY.html
- Versioning:    https://agent-manifest-spec.org/VERSIONING_POLICY.html
- Stability:     https://agent-manifest-spec.org/STABILITY.html
- Misconceptions: https://agent-manifest-spec.org/docs/MISCONCEPTIONS.html
- Adjacent specs: https://agent-manifest-spec.org/docs/comparison.html
- Foundations:   https://agent-manifest-spec.org/foundations/
- Public comment: https://agent-manifest-spec.org/nist-comment/nist-comment-2026.html

## Source repositories

- Organization:        https://github.com/agent-manifest
- Core specification:  https://github.com/agent-manifest/agent-manifest
- Public dataset:      https://github.com/agent-manifest/agent-manifest-dataset
- Registry governance: https://github.com/agent-manifest/agent-manifest-registry
- Ambassador (UI):     https://github.com/agent-manifest/agent-manifest-ambassador
- Diplomat (API):      https://github.com/agent-manifest/agent-manifest-diplomat
- Boundary Handshake:  https://github.com/agent-manifest/boundary-handshake
- Command-line interface (CLI): https://github.com/agent-manifest/agent-manifest-cli
- ∈ Principle:         https://github.com/agent-manifest/e-principle
- Agentic surface research: https://github.com/agent-manifest/agentic-surface-research
- Ambassador (live):   https://agent-manifest.github.io/agent-manifest-ambassador/

## Identifiers

- DOI:     https://doi.org/10.5281/zenodo.18833956
- Version: 1.0
- License: CC BY 4.0 (specification and documentation)
- Author:  Hernán Alfredo Capucci
- ORCID:   https://orcid.org/0009-0008-7216-3032
- Contact: contact@agent-manifest-spec.org
           (general; vulnerabilities go through the security policy instead)

## Command-line interface (CLI)

"CLI" here means "command-line interface". The Agent Manifest CLI is the
command-line validator for Agent Manifest v1.0 declarations.

- Name:          Agent Manifest command-line interface (CLI)
- Function:      command-line validator; checks a declaration against the v1.0
                 JSON Schema. Structural validation only.
- npm package:   @agent-manifest/cli
- Command:       agent-manifest
- Only command:  validate
- Version:       0.1.0
- Licence:       Apache-2.0
- Documentation: https://agent-manifest-spec.org/docs/cli/  (canonical page)
- Repository:    https://github.com/agent-manifest/agent-manifest-cli
- Registry:      https://www.npmjs.com/package/@agent-manifest/cli
- Install:       npm install -g @agent-manifest/cli
- Not:           it does not certify, score, enforce, or execute anything, and it
                 does not define the specification — it consumes it.

## Core doctrine

1. Declaration before execution.
   An agent that declares its identity, purpose, and boundaries before
   acting is structurally different from one that does not.

2. Defines boundaries, not behavior.
   The specification standardizes how agents declare what they are and
   what they will not do. It does not specify how agents execute.

3. Declaration Layer only.
   Agent Manifest sits below enforcement and execution layers. Validators,
   policy engines, audit systems, and runtimes are separate concerns.

4. Execution-agnostic.
   The specification is independent of any framework, runtime, language,
   or hosting environment.

5. Specification first. Product later.
   Adoption is voluntary. The specification is open, versioned, and
   designed to be composed with governance, discovery, and policy layers
   built by others.

## Field surface (v1.0)

A v1.0 Agent Manifest declares:

- agent_id, agent_name, agent_version
- manifest_version
- owner.type, owner.identifier
- purpose.primary_code, purpose.description
- forbidden_actions
- autonomy.level (0–3)
- risk_profile.level (low | medium | high)
- data_handling.stores_personal_data, data_handling.retention
- stopping_authority.stoppable_by, stopping_authority.mechanism
- audit_surface.logging, audit_surface.reconstructability
- contact.email

## Operational pipeline

  Ambassador  →  Diplomat  →  Dataset  →  Registry  →  Discovery

- Ambassador: conversational generator that produces v1.0 JSON manifests.
- Diplomat:   serverless registration endpoint that validates submissions.
- Dataset:    public, append-only archive of declared manifests.
- Registry:   governance and indexing of declarations.
- Discovery:  the .well-known JSON endpoint that points consumers to the
              registry and dataset.

## What Agent Manifest is not

- Not a runtime, framework, or agent execution engine.
- Not a behavioral monitor, observability layer, or audit system.
- Not a compliance certification or scoring service.
- Not a guarantee of behavior — a declaration of constraints, not a warrant
  that they are honoured.
- Not a replacement for governance, policy, or legal frameworks.

## Machine-readable representations

Each published Work co-locates its metadata with its landing page. For a Work at
https://agent-manifest-spec.org/works/<slug>/ the following are served:

- index.json        record summary
- schema.json       JSON-LD (schema.org)
- cite.json         CSL-JSON
- cite.bib          BibTeX
- cite.ris          RIS
- cite.md           Markdown citation
- cite-apa.txt      APA, plain text
- cite-plain.txt    plain-text citation
- dublin-core.json  Dublin Core
- highwire.json     Highwire citation_* tags (textual Works only)
- opengraph.json    OpenGraph
- signposting.json  FAIR Signposting link set

## Citation

Capucci, H. A. (2026). Agent Manifest v1.0. Zenodo.
https://doi.org/10.5281/zenodo.18833956

## Site credits

- Website design and development: OwnSite Studio (https://ownsitestudio.com)
- This credit covers the website only. The Agent Manifest specification, the Works and all research content are authored by Hernán Alfredo Capucci (ORCID 0009-0008-7216-3032) and cited via their own DOIs.