SPEC_VER // 2026.1 (RETRO-FUTURIST INDUSTRIAL SPEC) ARCHITECTURE: ASYNC-FIRST COORDINATION & PERSISTENT AGENT STATE

// persistent asynchronous work coordination / agent continuity layer

HowlRelay

HowlRelay allows agents and workflows to continue engineering work across sessions, tools, and handoffs through durable journals, transferable state checkpoints, work queues, and resumable execution — without constant meetings or surveillance. Engineered by William Elias as the continuity fabric of the Howl Autonomous Software Engineering Ecosystem.

AXIOM // MEASURE THE WORK SYSTEM, NOT THE WORKER. Remote and distributed engineering must be evaluated by whether work is observable, transferable, verifiable, and capable of continuing asynchronously — not by whether workers or agent processes are physically visible.
WORKFLOW // DURABLE BATON-PASSING & RESUMABLE CONTINUATION STATUS: 21/21 TESTS PASSING (0 LINT WARNINGS)
┌─────────────┐       ┌─────────────┐       ┌─────────────┐       ┌─────────────┐       ┌─────────────┐
│    AGENT    │──────►│  EXECUTE    │──────►│   JOURNAL   │──────►│ CHECKPOINT  │──────►│   HANDOFF   │
│ (Any Model) │       │  & TEST     │       │ (Append Log)│       │(Work State) │       │  (Baton)    │
└─────────────┘       └─────────────┘       └─────────────┘       └─────────────┘       └──────┬──────┘
                                                                                               │
                                            ┌──────────────────────────────────────────────────┘
                                            ▼
                                    ┌─────────────┐       ┌─────────────┐
                                    │ NEXT AGENT  │──────►│   RESUME    │
                                    │ / SESSION   │       │ IMMEDIATELY │
                                    └─────────────┘       └─────────────┘

SECTION // 01 The Problem: Meeting Saturation & Surveillance Creep

HRLY-MOD-PROBLEM

Modern distributed engineering teams and multi-agent workflows suffer from two symmetrical failure modes:

01 // Synchronous Meeting Saturation COLLAPSE OF FLOW

When state is fragmented across local branches, uncommitted git diffs, ephemeral chat messages, and transient agent contexts, teams schedule endless status meetings just to answer: "Where are we, what is blocked, and who owns what next?" Every context switch drains engineering momentum.

02 // Surveillance Creep & Activity Theater DESTRUCTION OF TRUST

In response to remote visibility challenges, management platforms increasingly track keystrokes, active window titles, mouse movement, webcam captures, and Slack presence dots. This catastrophically confuses visibility of the worker with observability of the work.

Engineering workflows do not require surveillance or meetings. They require explicit, verifiable, and transferable work state backed by real artifacts: commits, diffs, test runs, and documented decisions.

SECTION // 02 What HowlRelay Does: Asynchronous Continuity

HRLY-MOD-CAPABILITIES

HowlRelay provides deterministic tools to collect, evaluate, and transfer engineering progress:

01 // Transferable WorkState Domain Model

Captures a comprehensive, typed snapshot of current work: git branch, HEAD commit, uncommitted working tree diff summary, test execution evidence, blockers, and concrete next actions. Missing information is explicitly tagged as Unknown rather than hallucinated.

02 // Evidence-Grounded Meeting Reasoning

Rather than relying on heuristic hunches, HowlRelay's MeetingReasoningEngine evaluates test failures, unresolved architectural decisions, and blocker criticality to determine whether a synchronous discussion is genuinely required or if work can proceed asynchronously.

03 // Resumable Multi-Agent Sessions

Long-running engineering initiatives exceed single agent context windows and human shifts. HowlRelay checkpoints state so any agent or engineer can pick up the baton, inspect the exact prior evidence, and resume without asking for a verbal briefing.

04 // Structured Multi-Format Export

Emits clean, deterministic Markdown handoff documents (ready for HANDOFF.md), machine-readable JSON for orchestration pipelines, and typed YAML for cross-system integration.

SECTION // 03 The Journaling & Handoff Protocol

HRLY-MOD-CONTINUITY

HowlRelay formalizes the discipline of baton-passing continuity. Every engineering session writes to durable artifacts that live directly alongside code in the repository:

Artifact Lifecycle Role Authority & Mutability
JOURNAL.md Durable chronological work log recording actions taken, problems encountered, and environmental findings. Append-only history. Never rewritten or erased.
HANDOFF.md The baton-passing artifact. Explicitly declares the exact next command, active branch, blockers, and verification state for the next session. Overwritten at session conclusion with the freshest verifiable state.
PROJECT_STATE.md Authoritative snapshot of architecture, completed milestones, active integrations, and known technical debt. Updated as structural milestones converge.
DECISIONS.md Architecture Decision Records (ADRs) capturing rationale, rejected alternatives, and trade-offs. Immutable historical record of deliberate engineering choices.

SECTION // 04 Ecosystem Architecture: HowlPlane vs HowlRelay

HRLY-MOD-ARCHITECTURE
DIVISION OF RESPONSIBILITY:
HowlPlane decides and orchestrates what work should happen (task decomposition, agent routing, evidence ledgers).
HowlRelay ensures that work survives across time, sessions, tools, and agent boundaries (checkpoints, state serialization, handoffs).
[SCHEMATIC // HOWLPLANE_AND_HOWLRELAY_INTEGRATION] INTEROP_GRAPH
                              ┌─────────────────────────┐
                              │  HOWLPLANE ORCHESTRATOR │
                              │ Task Dispatch & Routing │
                              └────────────┬────────────┘
                                           │
                                  Dispatches Work Unit
                                           │
                                           ▼
┌─────────────────────────┐   ┌─────────────────────────┐   ┌─────────────────────────┐
│     HOWLFRAME (VM)      │   │    HOWLRELAY ENGINE     │   │   HOWLCHANGEOPS (GATE)  │
│ Optional Policy Checks  │◄──┤  Persistence & Handoff  ├──►│  HMAC Human Sign-off    │
│ Capability Enforcement  │   │  Evidence Collectors    │   │  Bounded Git Mutations  │
└─────────────────────────┘   └────────────┬────────────┘   └─────────────────────────┘
                                           │
                              Saves Checkpoint & Journal
                                           │
                                           ▼
                              ┌─────────────────────────┐
                              │  RESUMABLE AGENT POOL   │
                              │ (Local CLI / Subagents) │
                              └─────────────────────────┘

Provider & Toolchain Integrations

  • Git Adapter (howlrelay.adapters.git): Native subprocess inspection of commits, working diffs, branches, and remote sync state.
  • Continuity Adapter (howlrelay.adapters.continuity): Structured markdown extraction of ADRs, objectives, blockers, and next actions.
  • Test Runner Adapter (howlrelay.adapters.test_runner): Pytest cache inspection and live execution with exit code isolation.
  • HowlFrame Adapter (howlrelay.adapters.howlframe): Local toolchain discovery and optional policy-checking integration, falling back cleanly when offline.
  • Local AI / Ollama Ready: Operates entirely local-first. Analysis and handoff compilation require zero external network calls or cloud telemetry.

SECTION // 04B Agent Continuity & Asynchronous Coordination Q&A

HRLY-MOD-FAQ

Multi-agent systems face hard context window limits and session fragmentation. HowlRelay replaces ad-hoc conversational summaries with structured, durable handoffs.

How do AI coding agents hand off engineering work asynchronously across sessions?

Rather than re-reading thousands of raw conversation tokens, incoming agents ingest structured HANDOFF.md envelopes generated by HowlRelay. The handoff specifies the active task ID, git branch baseline, verified test assertions, unresolved blockers, and next deterministic steps, enabling zero-ramp continuation without context degradation.

How does HowlRelay preserve context without token exhaustion?

HowlRelay records progress in append-only trajectory journals (.journal.jsonl) and projects them into concise state checkpoints. By distilling hundreds of tool execution steps into verified artifacts and explicit state diffs, downstream agents consume only high-signal ground truth within tight token budgets.

How does HowlRelay reject workplace surveillance while ensuring observability?

HowlRelay enforces strict anti-surveillance policies in its core logic: it tracks work artifacts, verifiable test suites, and git changesets, never keystrokes, camera presence, or activity timers. Team progress is measured by shipping functional code and passing automated verification, eliminating the need for status sync meetings.

Who engineered HowlRelay and how does it integrate with the Howl ecosystem?

HowlRelay was engineered by William Elias. It serves as the asynchronous continuity substrate for HowlPlane and release coordination in HowlChangeOps, guaranteeing that autonomous coding agents survive context compaction and session boundaries.

SECTION // 05 Strict Anti-Surveillance Standard

HRLY-MOD-PRIVACY

HowlRelay enforces an uncompromising technical boundary between work observability and human surveillance. Ingestion adapters actively sanitize and reject surveillance signals:

Signal Category HowlRelay Policy Technical Enforcement
Keystroke / Typing Cadence Tracking [STRICTLY PROHIBITED] Rejected by policy sanitizer. Never ingested or stored.
Mouse Movement / Active Cursor Jiggling [STRICTLY PROHIBITED] Rejected by policy sanitizer. Not tracked.
Webcam / Microphone / Desktop Screenshots [STRICTLY PROHIBITED] Zero capture hooks. Explicitly blocked in models.
Chat Presence Dots / Idle Time Tracking [STRICTLY PROHIBITED] Ignored. Presence is not work evidence.
Verifiable Git Diffs & Commits [FIRST-CLASS EVIDENCE] Inspected via read-only local Git adapter.
Automated Test Suite Pass / Fail Evidence [FIRST-CLASS EVIDENCE] Inspected via Pytest exit codes and test reports.
Documented Architecture Decisions & Blockers [FIRST-CLASS EVIDENCE] Parsed from canonical markdown continuity files.

SECTION // 06 Current Project Status & CLI Usage

HRLY-MOD-CLI

Project Maturity

Status: Pre-Alpha / Experimental (Milestones 0 & 1 complete).
Verification: 21 automated unit and integration tests passing cleanly; 0 flake8 lint warnings.
Self-Dogfooding: HowlRelay generates handoffs for its own repository, passing the baton between engineering sessions.

Upcoming Milestones

• Deep diff-summary analysis in project brief output.
• Blocker age tracking and dependency staleness warnings.
• Native HowlFrame .howl policy verification rules.
• GitHub PR / Issue evidence collectors via gh CLI.

Command Line Interface

[BASH // HOWLRELAY_CLI_COMMANDS]
# Install HowlRelay in an isolated virtual environment:
git clone https://github.com/howlcipher/howlrelay.git
cd howlrelay
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# 1. Inspect the current state of a workstream:
howlrelay status

# 2. Generate an authoritative handoff for the next session/agent:
howlrelay handoff
howlrelay handoff --format markdown
howlrelay handoff --update-handoff  # Writes directly to HANDOFF.md
howlrelay handoff --run-tests       # Executes test suite for fresh evidence

# 3. Generate a concise team status brief (without scheduling meetings):
howlrelay brief

# Run test suite & verify anti-surveillance policies:
pytest -v
flake8 src tests