Intent building Intent

Every artifact on this site was created using the methodology it describes. This page shows the live state of Intent's own signal stream, specifications, and event log — proof the system works by working on itself.

Signals captured

Spec documents

Decisions

MCP servers

Knowledge artifacts

The meta-loop

Intent's development follows its own Notice → Spec → Execute → Observe loop. Observations about how work flows (or doesn't) get captured as signals. Signals cluster into problems worth solving. Specs define what to build. Agents execute. Events get emitted. And the loop continues.

What you're looking at below is the actual output of that loop — not mock data, not a demo. These are real signals from real sessions, feeding real specs, generating real events. Knowledge Engine (Layer 1) feeds the Intent loop through lint-generated signals and knowledge queries during spec-shaping.

24 Signals

→

14 Specs

→

12 Pages

→

∞ Events

Signal Stream

Live

SIG-001 Intent needs a formal work ontology Mar 28 SIG-002 Distributed tracing is the right observability model for Intent work Mar 28 SIG-003 Conversations should automatically feed the notice layer Mar 28 SIG-004 Intent needs three governance dimensions, not just the loop Mar 28 SIG-005 Work units need formal schemas to be agent-consumable Mar 28 SIG-006 Intent is a description, not an enabler — practitioners can understand it but can't install and test it Monday morning Mar 29 SIG-007 Teams using AI agents hit a ceremony wall around sprint 3 Mar 29 SIG-008 Signals die in the gap between where they're noticed and where the system can see them Mar 29 SIG-009 Intent is four products (Notice, Spec, Execute, Observe), not one — each needs its own roadmap Mar 29 SIG-010 Engineer's team rewired around AI: tickets became bot specs, refinement became design sessions Mar 29 SIG-011 Signal capture needs to cover every surface where practitioners work — not just the IDE Mar 29 SIG-012 Autonomous signal processing with trust-based execution levels Mar 29 SIG-013 Multi-machine cloud requirement: can't rely on local laptop being always connected Mar 29 SIG-014 Agent context limits cause content drift during multi-file pushes Mar 29 SIG-015 Signal amplification through reference frequency Mar 29 SIG-016 Intent needs a multi-agent MCP architecture Mar 30 SIG-017 OTel deployment is a spectrum, not a binary choice Mar 30 SIG-018 Free-tier cloud MCP hosting is viable for Intent's agent layer Mar 30 SIG-019 "Notice" and "Execute" aren't landing as loop phase names Mar 30 SIG-020 Intent site needs two-tier information architecture Mar 30 SIG-021 Working MCP server code exists for all three loop phases Mar 30 SIG-022 Sequential signal IDs will collide in distributed multi-agent environments Mar 30 SIG-023 Signals need cross-team portability with provenance-carrying IDs Mar 30 SIG-024 Cowork sandbox cannot reliably sync to GitHub — needs Claude Code CLI path Mar 30

Specification Documents

14 specs

Intent Methodology

The full spec: loop, toolchain, and what replaces what

Concept Brief

Problem, audience, solution, GTM, hypotheses

Work Ontology

Two-plane architecture: flow + ownership

Signal Capture System

CLI, MCP, and multi-surface capture design

Trust Framework

L0-L4 autonomy levels and trust scoring

Signal Amplification

Reference-weighted, time-decayed signal strength

Event Catalog

15 event types across signal, intent, spec, contract

Flow Diagram

Visual pipeline from signal to deployed artifact

Signal Stream

Dashboard design and signal lifecycle

Product Roadmap

Four-product investment strategy

Repo Pattern

Intent-native repository structure

Decision Log

Institutional memory: decisions with rationale

Autonomous Operations

Three-layer migration: sandbox → CLI → cloud

Event Stream

events.jsonl

Mar 28signal.createdSIG-001 work-ontology

Mar 28signal.createdSIG-002 otel-for-work

Mar 28signal.createdSIG-003 three-dimensions

Mar 28signal.createdSIG-004 units-need-schemas

Mar 28signal.createdSIG-005 conversation-as-signal

Mar 28system.statusv0.5.0 site launched

Mar 29signal.createdSIG-006 four-products-not-one

Mar 29signal.createdSIG-007 ceremony-wall

Mar 29signal.createdSIG-008 ari-pattern

Mar 29signal.createdSIG-009 signals-die-in-context-switch

Mar 29signal.createdSIG-010 signal-amplification

Mar 29signal.createdSIG-011 bootstrap-gap

Mar 29signal.createdSIG-012 autonomous-processing

Mar 29signal.createdSIG-013 multi-surface-capture

Mar 29signal.createdSIG-014 always-on-requirement

Mar 29signal.createdSIG-015 agent-context-limits

Mar 29system.statusv0.7.0 CLI tools shipped

Mar 30spec.approvedwork-ontology v2 (two-plane)

Mar 30system.statusv0.8.0 two-plane architecture, pitch page, dogfood

Why this matters

Most operating models are described in slide decks. Intent is described in its own format. Every signal on this page was captured using intent-signal. Every spec was written using the Intent template. Every event was emitted by the Intent GitHub Action.

This isn't just documentation. It's a live system observing itself. When we notice friction in using Intent to build Intent — that observation becomes a signal, which feeds the next iteration. The methodology improves by running.

The methodology this system follows →