monarchic/agent-protocol

Monarchic Agent Protocol protobuf bindings for PHP.

Maintainers

Package info

github.com/monarchic-ai/monarchic-agent-protocol

Language:C#

pkg:composer/monarchic/agent-protocol

Statistics

Installs: 0

Dependents: 0

Suggesters: 0

Stars: 0

Open Issues: 1

Security

Advisories: 0

Aikido package health analysis

0.1.16 2026-04-01 18:06 UTC

Requires

LGPL-3.0-only 34b277b5177424bdca2836ee6e0eeb952ee0cff7

This package is auto-updated.

Last update: 2026-04-01 18:07:51 UTC

README

ci release

This repository defines the shared, versioned protocol for Monarchic AI. It is the compatibility layer between the orchestrator, runner, and agent roles, so the schemas are minimal and stable while allowing forward-compatible extensions.

Overview

  • Provide versioned JSON Schemas for language-agnostic validation.
  • Provide Rust, TypeScript, and Protobuf bindings that mirror the schemas.
  • Keep the protocol small and explicit for v1 interoperability.

Usage

Quickstart

Install the published package for your language, then use the generated bindings.

Install

  • Rust (crates.io): cargo add monarchic-agent-protocol
  • TypeScript (npm): npm install @monarchic-ai/monarchic-agent-protocol
  • Python (PyPI): pip install monarchic-agent-protocol
  • Ruby (RubyGems): gem install monarchic-agent-protocol
  • Go (Go modules): go get github.com/monarchic-ai/monarchic-agent-protocol/src/go@vX.Y.Z
  • Java/Kotlin (JitPack): implementation("com.github.monarchic-ai:monarchic-agent-protocol:vX.Y.Z")
  • .NET (NuGet): dotnet add package Monarchic.AgentProtocol
  • PHP (Packagist): composer require monarchic-ai/monarchic-agent-protocol

Examples

  • Rust: examples/rust/task.rs
  • TypeScript: examples/ts/task.ts
  • JSON (non-protobuf): examples/json/objective_spec.minimal.json
  • Protobuf C++: examples/proto/cpp/task.cpp
  • Protobuf Java: examples/proto/java/TaskExample.java
  • Protobuf Kotlin: examples/proto/kotlin/TaskExample.kt
  • Protobuf C#: examples/proto/csharp/TaskExample.cs
  • Protobuf Python: examples/proto/python/task.py
  • Protobuf Ruby: examples/proto/ruby/task.rb
  • Protobuf PHP: examples/proto/php/task.php
  • Protobuf Dart: examples/proto/dart/task.dart
  • Protobuf Rust: examples/proto/rust/task.rs

Non-protobuf language support

For languages that do not use protobuf bindings, exchange protocol objects as JSON and validate payloads against the versioned schemas before handoff.

  • Canonical JSON example for non-protobuf consumers: examples/json/objective_spec.minimal.json
  • Validate a typed payload against a specific schema: bash scripts/validate-protocol-json.sh schemas/v1/objective_spec.json schemas/fixtures/valid/objective_spec.minimal.json
  • Validate the canonical non-protobuf JSON example: bash scripts/validate-protocol-json.sh schemas/v1/objective_spec.json examples/json/objective_spec.minimal.json

Project-state contract fixtures used by the shell verification suite live under fixtures/project-state/, not the repo root.

Versioning

  • Protocol versions live under schemas/v1/.
  • Each v1 object requires version: "v1".
  • New versions must be added under a new directory (e.g. schemas/v2/) without changing existing v1 files.

Client boundary contracts

The Rust crate also exposes a frozen client-boundary surface under monarchic_agent_protocol::client_boundary.

Frozen v1 client-boundary types:

  • Intent
  • Plan
  • PlanStep
  • ExecutionReceipt
  • VerificationReceipt
  • ReviewDecision
  • RerunScope
  • BlockedOutcome
  • ArtifactDescriptor

Canonical JSON fixtures for these contracts live under fixtures/client_boundary/v1/. Legacy runtime compatibility fixtures live under fixtures/client_boundary/v0/.

Durable authority contracts

The Rust crate also exposes a frozen durable-authority surface under monarchic_agent_protocol::durable_authority.

Frozen v1 durable-authority types:

  • FencingToken
  • Lease
  • RunLifecycleState
  • StepLifecycleState
  • LeaseRejectionReason
  • RecoveryEvent

Canonical JSON fixtures for these contracts live under fixtures/durable_authority/v1/.

Schema summary

JSON Schema files are generated from the protobuf definitions. Do not edit them by hand.

Schema files live under schemas/v1/:

  • schemas/v1/task.json
  • schemas/v1/artifact.json
  • schemas/v1/event.json
  • schemas/v1/gate_result.json
  • schemas/v1/failure_class.json
  • schemas/v1/plan_status.json
  • schemas/v1/failure_detail.json
  • schemas/v1/role_provenance.json
  • schemas/v1/plan_provenance.json
  • schemas/v1/plan_step.json
  • schemas/v1/plan.json
  • schemas/v1/execution_receipt.json
  • schemas/v1/run_context.json
  • schemas/v1/run_outcome.json
  • schemas/v1/delivery_contract.json
  • schemas/v1/agent_role.json
  • schemas/v1/dataset_ref.json
  • schemas/v1/experiment_spec.json
  • schemas/v1/objective_spec.json
  • schemas/v1/eval_result.json
  • schemas/v1/provenance.json
  • schemas/v1/schema.json (index)
  • schemas/v1/monarchic_agent_protocol.proto

All schemas allow additional properties for forward compatibility.

Schema index coverage

schemas/v1/schema.json is the canonical top-level JSON schema index. Its oneOf entries currently cover:

  • schemas/v1/task.json
  • schemas/v1/artifact.json
  • schemas/v1/event.json
  • schemas/v1/gate_result.json
  • schemas/v1/failure_class.json
  • schemas/v1/plan_status.json
  • schemas/v1/failure_detail.json
  • schemas/v1/role_provenance.json
  • schemas/v1/plan_provenance.json
  • schemas/v1/plan_step.json
  • schemas/v1/plan.json
  • schemas/v1/execution_receipt.json
  • schemas/v1/run_context.json
  • schemas/v1/dataset_ref.json
  • schemas/v1/experiment_spec.json
  • schemas/v1/objective_spec.json
  • schemas/v1/eval_result.json
  • schemas/v1/provenance.json

schemas/v1/agent_role.json is a shared schema used by task.json. schemas/v1/failure_class.json is a shared schema used by event.json and gate_result.json.

TaskMessage and TaskMessageAck

These types define the shared contract for orchestrator-mediated runner communication.

They are intended for:

  • durable handoff messages between active tasks
  • clarification requests and responses
  • blocker notices
  • artifact-ready notifications
  • explicit acknowledgement state

They are not intended to imply direct peer-to-peer runner transport. The protocol defines the message shape, but routing, persistence, and delivery are owned by the orchestrator.

TaskMessage carries:

  • sender and recipient task ids
  • message kind
  • optional subject/body
  • referenced artifact ids
  • optional reply chaining
  • acknowledgement requirement

TaskMessageAck records recipient acknowledgement state separately so mailbox delivery can remain append-only and auditable.

Recommended acknowledgement semantics:

  • received: the recipient has seen the message in its inbox
  • accepted: the recipient accepts the request and plans to act on it
  • rejected: the recipient explicitly declines or cannot act on it
  • resolved: the recipient completed the requested follow-up or supplied the final response

Recommended routing semantics:

  • message ids should be unique within one run
  • sender and recipient should be tasks from the same run
  • reply_to, when present, should reference an earlier message id from the same run log
  • requires_ack=true should imply at least one corresponding TaskMessageAck record from the recipient

AgentRole

Enum values:

  • product_owner
  • project_manager
  • dev
  • qa
  • reviewer
  • security
  • ops
  • publisher

Example:

{
  "role": "reviewer"
}

PipelineSpec

Represents a planned pipeline before execution.

Required fields:

  • version: "v1"
  • pipeline_id: stable identifier
  • objective: human-readable campaign or pipeline objective
  • project_key: member/project scope identifier
  • tasks: ordered PipelineTask[]

This is the shared planning shape that bootstrap generation, orchestration validation, and UI preview should converge on.

Current shared planning fields now include:

  • PipelineSpec
  • PipelineTask
  • TaskDependency
  • SkillRef
  • RoleDefinition
  • ResolvedRoleBundle

These are available in the protobuf and language bindings even where the checked-in JSON Schema index has not yet been expanded to cover each planning helper type.

RoleDefinition and ResolvedRoleBundle

These types provide the shared contract between:

  • role catalogs in monarchic-agent-roles
  • orchestration-time validation
  • runner-time execution bundles

RoleDefinition describes a canonical role, its capabilities, and its declared skill requirements. ResolvedRoleBundle is the runtime handoff shape that can pair a concrete role definition with the resolved skills and rendered template path used for one task execution.

Canonical Pipeline Layout

PipelineSpec and PipelineTask are now the canonical role-aware planning layout across the stack.

For compatibility, older minimal pipeline files may still only carry:

  • pipeline_id
  • tasks[].id
  • tasks[].task

But once a pipeline opts into the role-aware contract by declaring any of:

  • objective
  • project_key
  • tasks[].role
  • tasks[].goal
  • tasks[].required_skills

the intended canonical shape is:

  • PipelineSpec.objective: required, non-empty
  • PipelineSpec.project_key: required, non-empty
  • PipelineTask.role: required, non-empty
  • PipelineTask.goal: required, non-empty
  • PipelineTask.required_skills: optional list of SkillRef, but when present it must be internally well-formed and deduplicated

This is the contract monarch should generate, monarchic-orchestrator should validate, and monarchic-runner should ultimately execute through resolved role bundles.

Task

Represents work assigned to an agent.

Required fields:

  • version: "v1"
  • task_id: stable identifier
  • role: AgentRole
  • goal: human-readable objective

Optional fields:

  • inputs: free-form object
  • constraints: free-form object
  • gates_required: list of gate names to run (ex: ["qa", "security"])
  • run_context: RunContext
  • objective_spec: objective scoring contract for deterministic outcome evaluation
  • experiment_spec: typed experiment design contract for deterministic in silico runs

Example:

{
  "version": "v1",
  "task_id": "task-123",
  "role": "dev",
  "goal": "Implement protocol types",
  "inputs": {
    "issue": "https://example.com/issues/42"
  },
  "constraints": {
    "no_network": true
  },
  "gates_required": ["qa", "security"],
  "run_context": {
    "version": "v1",
    "repo": "monarchic-agent-protocol",
    "worktree": "/worktrees/task-123",
    "image": "ghcr.io/monarchic/runner:stable",
    "runner": "vm-runner-01",
    "labels": ["linux", "rust"]
  }
}

DeliveryContract

Typed acceptance contract for autonomous delivery execution.

Required fields:

  • objective: plain-language objective statement
  • definition_of_done: ordered completion checklist
  • required_checks: required gate/check names for merge readiness
  • risk_tier: low | medium | high | critical

Optional fields include cycle/turn budgets (max_cycle_minutes, max_agent_turns) and PR/review/rollback strategy hints.

RunContext

Execution hints for a runner.

Required fields:

  • version: "v1"
  • repo: repository identifier or URL
  • worktree: worktree path or identifier
  • image: VM/container image reference
  • runner: runner identifier

Optional fields:

  • labels: list of labels or tags

Example:

{
  "version": "v1",
  "repo": "monarchic-agent-protocol",
  "worktree": "/worktrees/task-123",
  "image": "ghcr.io/monarchic/runner:stable",
  "runner": "vm-runner-01",
  "labels": ["linux", "rust"]
}

Artifact

Outputs produced by an agent or runner.

Required fields:

  • version: "v1"
  • artifact_id: stable identifier
  • type: artifact type (ex: patch, log, report)
  • summary: short description
  • path: path or locator for the artifact
  • task_id: task identifier that produced it
  • provenance: typed provenance hashes and runtime metadata
  • dataset_refs: datasets used while producing the artifact
  • eval_results: typed metric outputs with optional uncertainty bounds
  • experiment_spec: optional copy of experiment contract used for this output

Example:

{
  "version": "v1",
  "artifact_id": "artifact-987",
  "type": "patch",
  "summary": "Adds v1 protocol schemas",
  "path": "artifacts/task-123/patch.diff",
  "task_id": "task-123"
}

Event

Lifecycle state updates.

Required fields:

  • version: "v1"
  • event_type: event category
  • timestamp: ISO 8601 timestamp
  • task_id: task identifier
  • status: state label

Optional fields:

  • message: human-readable details
  • provenance: typed runtime/source hashes for event attribution
  • eval_results: optional metric snapshot payloads
  • failure_class: typed failure taxonomy payload for machine-actionable triage

Example:

{
  "version": "v1",
  "event_type": "task_started",
  "timestamp": "2025-01-14T15:04:05Z",
  "task_id": "task-123",
  "status": "running",
  "message": "Runner started VM"
}

GateResult

Outcome of QA, review, security, or other gates.

Required fields:

  • version: "v1"
  • gate: gate name
  • status: pass/fail or other gate-specific status

Optional fields:

  • reason: short explanation
  • failure_class: typed failure taxonomy payload for deterministic failure routing
  • evidence: free-form object with supporting data

Example:

{
  "version": "v1",
  "gate": "security",
  "status": "pass",
  "reason": "No high or critical findings",
  "evidence": {
    "scanner": "trivy",
    "report_path": "artifacts/task-123/security.json"
  }
}

FailureClass

Typed taxonomy payload for classifying protocol failures.

Required fields:

  • category: validation, dependency, environment, timeout, conflict, permission, resource, internal, or unknown
  • code: stable machine-readable failure code
  • retryable: whether automated retry is expected to be useful

Optional fields include detail, scope, source, and next_action.

PlanStatus

Plan lifecycle status used for typed execution plans and receipts.

Allowed values:

  • unspecified
  • draft
  • planned
  • executing
  • complete
  • bounded
  • failed
  • cancelled
  • unknown

FailureDetail

Failure detail attached to a plan or receipt.

Required fields:

  • class: one of validation, execution, agent, infra, policy, or unknown
  • code: machine-readable failure code
  • message: operator-readable failure message

Optional fields:

  • details: bounded extension object with implementation diagnostics

RoleProvenance

Role metadata for deterministic role-template binding.

Required fields:

  • role_name
  • template_hash
  • render_hash

PlanProvenance

Generation metadata for a plan and policy context.

Required fields:

  • generated_by
  • generated_at_ms

Optional fields:

  • policy_profile
  • role

PlanStep

Execution step in a plan, with dependency and template fields.

Required fields:

  • step_id
  • description
  • task_template

Optional fields:

  • depends_on
  • failure

Plan

Canonical plan contract.

Required fields:

  • contract_version: required contract tag (for now "v1")
  • plan_id
  • objective
  • status
  • created_at_ms
  • updated_at_ms
  • provenance
  • steps

Optional fields:

  • run_id

ExecutionReceipt

Deterministic execution contract for a run.

Required fields:

  • contract_version: required contract tag (for now "v1")
  • run_id
  • plan_id
  • plan_hash
  • task_hashes
  • artifact_hashes
  • outcome_hash
  • status
  • generated_at_ms

Optional fields:

  • failure

DatasetRef

Reference to a dataset used by an experiment or output artifact.

Required fields:

  • dataset_id
  • sha256
  • format

Optional fields include uri, split, size_bytes, and description.

ExperimentSpec

Typed contract for model design and acceptance checks.

Required fields:

  • experiment_id
  • objective
  • dataset_refs: list of DatasetRef
  • acceptance: metric threshold policy

Optional fields include hypothesis, model_family, seeds, and free-form constraints.

ObjectiveSpec

Typed objective scoring contract for progress/outcome checks.

Required fields:

  • metric_key
  • direction: maximize, minimize, or target

Optional fields include target, min_delta, tolerance, report_file, report_task_id, weight, and description.

EvalResult

Typed evaluation output row.

Required fields:

  • metric
  • value
  • passed

Optional fields include lower_ci, upper_ci, variance, seed, and notes.

Provenance

Typed provenance contract for reproducibility and traceability.

Required fields:

  • prompt_sha256
  • code_sha256
  • runtime
  • runner
  • orchestrator
  • created_at

Optional fields include dataset hashes/references and command/task/pipeline hashes.

Rust

The crate lives at the repo root with sources under src/rust/lib.rs.

use monarchic_agent_protocol::{AgentRole, Task, PROTOCOL_VERSION};

let task = Task {
    version: PROTOCOL_VERSION.to_string(),
    task_id: "task-123".to_string(),
    role: AgentRole::Dev as i32,
    role_id: "dev".to_string(),
    goal: "Implement protocol".to_string(),
    inputs: None,
    constraints: None,
    gates_required: Vec::new(),
    run_context: None,
    extensions: Default::default(),
};

TypeScript

TypeScript bindings are in src/ts/index.ts.

import { Task } from "./src/ts/index";

const task: Task = {
  version: "v1",
  task_id: "task-123",
  role: "dev",
  goal: "Implement protocol",
};

Go

Go module sources live under src/go with module path:

github.com/monarchic-ai/monarchic-agent-protocol/src/go

Protobuf

The v1 protobuf schema lives at schemas/v1/monarchic_agent_protocol.proto. Its portable object messages mirror the JSON schema and use google.protobuf.Struct for free-form objects (inputs, constraints, evidence, extensions). Additional JSON properties should be stored in the extensions field on each message.

The protobuf file also contains transport-only service definitions for machine-to-machine control-plane RPC, starting with RunnerControlService. Those protobuf-only contracts cover registration, heartbeats, lease acquisition, lease renewal, lease resume, step progress/outcome reporting, and cancellation acknowledgement, and they are not mirrored into the JSON schema index.

The Rust crate published from this repository currently exports protobuf message types. Runtime repositories that need gRPC client/server bindings should generate transport stubs from the canonical proto with their chosen toolchain, such as tonic-build.

Language packages are published per registry. Use the registry package for your language instead of generating local outputs.

Python (PyPI)

Install the published package and import the generated protobuf bindings:

from monarchic_agent_protocol import monarchic_agent_protocol_pb2 as map_pb2

Ruby

Ruby bindings live under src/ruby.

Java/Kotlin

Java/Kotlin sources live under src/java.

C#

C# sources live under src/csharp.

PHP

PHP sources live under src/php.

Dart

Dart sources live under src/dart.

Contributing

Tooling

  • nix develop provides Rust, Node, jq, Python jsonschema, and protoc.
  • nix flake check validates JSON schemas, protobuf codegen, and package imports (PyPI + Rust + npm + Go).
  • JSON Schema test: scripts/test-json-schema.sh.
  • Language-agnostic schema validation helper: scripts/validate-protocol-json.sh.
  • Language-agnostic schema validator regression test: scripts/test-validate-protocol-json.sh.
  • Pre-commit schema JSON parse check: scripts/pre-commit-schema-json-parse.sh.
  • Pre-commit schema parse smoke test: scripts/test-pre-commit-schema-json-parse.sh.
  • Schema edit changelog: schemas/SCHEMA_CHANGELOG.md.
  • Schema changelog format test: scripts/test-schema-changelog-format.sh.
  • README schema index coverage test: scripts/test-readme-schema-index-coverage.sh.
  • README examples coverage test: scripts/test-readme-examples-coverage.sh.
  • README/examples examples synchronization test: scripts/test-readme-examples-sync.sh.
  • Protobuf codegen test (all languages): scripts/test-proto.sh.
  • Protobuf availability smoke test: scripts/test-proto-availability-smoke.sh.
  • Protobuf codegen (write to src/<lang>): scripts/generate-proto.sh.
  • Protobuf codegen (write to src/<lang> and regenerate JSON schemas): scripts/generate-proto.sh.
  • JSON Schema regeneration only: scripts/generate-json-schema.sh.
  • JSON Schema generation requires protoc-gen-jsonschema (install with go install github.com/chrusty/protoc-gen-jsonschema/cmd/protoc-gen-jsonschema@latest).

Schema validation workflow

  1. Run full schema lint and semantic checks: bash scripts/lint-schemas.sh.
  2. Run direct schema fixture checks: bash scripts/test-json-schema.sh.
  3. Validate language-agnostic payloads through JSON schema: bash scripts/validate-protocol-json.sh schemas/v1/objective_spec.json schemas/fixtures/valid/objective_spec.minimal.json.
  4. Verify schema validator behavior is deterministic: bash scripts/test-validate-protocol-json.sh.
  5. Validate staged schema JSON before commit: bash scripts/pre-commit-schema-json-parse.sh.
  6. Verify pre-commit checker behavior is deterministic: bash scripts/test-pre-commit-schema-json-parse.sh.
  7. Verify schema changelog entry format: bash scripts/test-schema-changelog-format.sh.
  8. Verify README schema index coverage stays aligned: bash scripts/test-readme-schema-index-coverage.sh.
  9. Verify README examples coverage for non-protobuf and protobuf paths: bash scripts/test-readme-examples-coverage.sh.
  10. Verify README and examples/README example entries stay synchronized: bash scripts/test-readme-examples-sync.sh.

Nix packages

  • packages.default: Rust crate for protocol types
  • packages.rs-lib: Rust crate for protocol types (local)
  • packages.rs-registry-lib: Rust crate from crates.io (registry)
  • packages.py-lib: installable Python package (local)
  • packages.py-registry-lib: PyPI package (registry)
  • packages.ts-lib: TypeScript types package (local)
  • packages.ts-registry-lib: npm registry package (types-only)
  • packages.go-lib: Go module (local)
  • packages.go-registry-lib: Go module from GitHub (registry)
  • packages.rb-lib: Ruby gem (local)
  • packages.rb-registry-lib: Ruby gem from RubyGems (registry)
  • packages.java-lib: Java/Kotlin jar (local)
  • packages.java-registry-lib: Java/Kotlin jar from JitPack (registry)
  • packages.dart-lib: Dart package (local)
  • packages.dart-registry-lib: Dart package from pub.dev (registry)
  • packages.csharp-lib: C# package sources (local)
  • packages.csharp-registry-lib: C# package from NuGet (registry)
  • packages.php-lib: PHP package sources (local)
  • packages.php-registry-lib: PHP package from Packagist (registry)

CI and releases

  • .github/workflows/ci.yml validates JSON schemas, protobuf codegen, and runs cargo test.
  • .github/workflows/release.yml publishes language packages.
    • Python publishing is implemented for PyPI; other language registry steps are scaffolded.

License

LGPL-3.0-only