Field Report 001

Relayfox: file-native coordination for mixed agents

Relayfox is the first useful proof object behind LeveragedMindHQ: a small, file-native task system for keeping AI-assisted work bounded, reviewable, and human-controlled.

The Problem

AI can make execution feel instant, but it also makes drift easier. A model can start with a clear goal and end with a pile of unreviewable changes, invented checks, or missing context. The problem is not intelligence. The problem is coordination.

What Was Built

Relayfox uses durable task packets instead of relying on chat memory. A planner writes the context and acceptance criteria into files. An executor claims a task, works inside the allowed scope, runs real checks, and records proof in a result file.

• document.md preserves working memory.
• brief.md defines the canonical task.
• handoff.md bounds the executor assignment.
• result.md records changed files, checks, uncertainty, and next action.

Proof In This Repository

The proof is deliberately mundane: scripts, tests, task folders, and lint output. Relayfox is not being presented here as a finished platform or autonomous dispatch network. It is a working coordination scaffold that makes lower-cost execution safer.

• relayfox/scripts/claim-task.py protects a task from multiple executors claiming it silently.
• relayfox/scripts/lint-task.py checks task packet structure.
• relayfox/scripts/verify-result.py exists to compare claimed checks against actual proof.
• relayfox/tests/ provides the current unit-test surface.

What Broke Or Remains Unfinished

Relayfox does not magically make agents reliable. It exposes the real cost: specification, review, and integration. It still needs better trace views, stronger result verification, and clearer boundaries between mission-level decisions and task-level execution.

The Model Learned

This is the Coordination Tax in practice. More agents can increase throughput, but only if the work is decomposed, claimed, checked, and reviewed. Otherwise, cheap execution becomes expensive cleanup.