Project Phoenix

Appearance

Seven Agents. One Mission. ​

Extract the intent. Rebuild from zero.

The Pipeline ​

Project Phoenix decomposes legacy modernization into seven sequential stages, each handled by a specialized AI agent. The output of each agent feeds the next. Human validation gates separate every stage.

The Agents ​

A-00: Signal Extraction ​

Project Context → Mission Brief

Before any agent touches the codebase, A-00 reads everything known about the system — docs, specs, schemas, stakeholder notes — and produces the mission brief. Every downstream agent reads it before beginning. If nothing exists, A-00 scans the system surface itself. The brief is orientation, not analysis.

Outputs: Mission brief (_mission.sil), signal .sil files — one per known workflow

Agent A-01: Business Logic Extractor ​

Source Code → Business Requirements

Reads the entire legacy codebase and extracts every business rule, workflow, decision tree, validation, calculation, and edge case. Produces a structured catalog of what the system actually does — not how it does it.

Outputs: Business rules catalog, workflow maps, data entity model, dependency graph, edge case registry

Agent A-02: Interface Archaeologist ​

UI/UX → User Intent Model

Maps every screen, form, report, and user interaction in the legacy system. Reconstructs what users are trying to accomplish at each step — the intent behind the interface.

Outputs: Screen inventory, user journey maps, input/output specs, role-permission matrix, report catalog

Agent A-03: Requirements Synthesizer ​

Logic + Interface → Unified Specification

Weaves the business logic from Agent 1 and the user intent from Agent 2 into a single, coherent specification. Identifies gaps, contradictions, and ambiguities that require human resolution.

Outputs: Unified requirements spec, logic-to-UI mapping, gap analysis, ambiguity flags

Agent A-04: Solution Architect ​

Requirements → Stack & Blueprint

Takes the unified specification and designs the optimal modern implementation. Selects the tech stack, defines API contracts, designs the data schema, and creates the build plan.

Outputs: Tech stack recommendation, system architecture, API contracts, data schema, build plan

Agent A-05: Builder Fleet ​

Blueprint → Working Software

A coordinated swarm of coding agents that build the greenfield application in parallel. Backend, frontend, database, integrations, CI/CD — all constructed simultaneously from the blueprint. No legacy code, no translation.

Outputs: Production codebase, deployment pipeline, documentation, infrastructure-as-code

Agent A-06: Validator & Certifier ​

New System ↔ Original Requirements

The critical final gate. Regression-tests the new system against the original business rules extracted by Agent 1. Proves that the new system does everything the old system did — and nothing it shouldn't.

Outputs: Test suite, regression results, coverage report, certification document, gap exceptions list

Key Principles ​

Sequential, Not Parallel ​

The agents run in sequence because each depends on the output of the previous stage. You can't architect a system before you have requirements. You can't build before you have a blueprint. You can't validate before you have a build.

Gated, Not Autonomous ​

Human validation gates between every stage prevent compounding errors. An extraction mistake caught early doesn't become an architectural flaw downstream.

Greenfield, Always ​

No agent reads legacy code to produce legacy-style output. Agent 5 builds from the blueprint — it never sees the original codebase. The new system is modern by construction, not by translation.

"The pipeline doesn't translate. It understands, then builds."

Artifact Format & Runtime ​

Every agent reads and writes .sil files — EMBER's structured, human-readable artifact format. One construct per file. Plain text. Git diffable. Permanent.

The pipeline is orchestrated by @semanticintent/phoenix-runtime — an open-source CLI that manages state, enforces human gates, and produces agent prompts ready to run in any AI interface.

bash
npm install @semanticintent/phoenix-runtime

phoenix init my-project
phoenix run a-00
phoenix status
phoenix gate pass-1 --approve