phoenix-runtime

Phoenix Pipeline Runtime — orchestrate the 7-agent legacy modernization pipeline.

Manages .sil artifact state, enforces human gates, injects episode context, and produces agent prompts ready to run in Claude Code or any AI interface.

phoenix init acme-order-system
phoenix run a-00
phoenix status
phoenix gate pass-1 --approve --notes "screens look correct"
phoenix episode new
phoenix run a-06

How It Works

The runtime does not call an AI API. It produces prompts. You paste them into Claude Code, Claude CLI, or any interface. This keeps the runtime lightweight, model-agnostic, and free of API key management.

State lives in .phoenix/state.json at your project root — human readable, git diffable, travels with the project.

The Pipeline

| Agent | Name | Produces | |-------|------|----------| | A-00 | Signal Extraction | Mission brief + signal files | | A-01 | Business Logic Extractor | ASCII workflow traces | | A-02 | UI Archaeologist | ASCII wireframes | | A-03 | Requirements Synthesizer | Semantic intent specs | | A-04 | Solution Architect | Stack recommendation + blueprints | | A-05 | Builder | Production codebase (6 passes) | | A-06 | Validator & Certifier | Certification document |

Human sign-off required between every pass in A-05, and before A-05 can begin.

First Engagement: wealth2track

Flutter/Firebase real estate portfolio app (2020) → React 18 + TypeScript web app (2026)
Live: w2t.semanticintent.ai

Source system: ~9,160 lines of Dart across 8 Firestore services, 14 screens, 40+ models. Built 2020, paused at beta. Firebase auth, Firestore, Google Maps Places API.

Pipeline run: April 3, 2026 — all 7 agents in a single session.

| Agent | Output | |-------|--------| | A-00 | 14 signal files across 5 domains — all services, routes, models read | | A-01 | 14 workflow traces — 4 bugs found including cascade delete targeting wrong collection, LTV division-by-zero | | A-02 | 15 screen wireframes — 8 UX findings flagged | | A-03 | 14 specs — 5 divergences, 8 confirmed bugs requiring decisions | | A-04 | 4 architecture files — React 18 + TypeScript, Tailwind, Zustand, Dexie.js, Vite; no auth, no Firebase | | A-05 | 5 build passes — 51 tests passing, zero TS errors, 407KB bundle | | A-06 | Certification — CONDITIONAL PASS, 9/14 certified, 5 with notes, 11 original bugs resolved in rebuild |

Episode mid-engagement: At the A-04 review gate, the decision was made not to modernize Flutter but to replatform to a local-first web app. One episode recorded — all downstream agents received the context automatically, no re-runs needed.

phoenix episode new
# → ep-001: Stack pivot — Proton-like web app, Dexie.js/IndexedDB, no Firebase, no auth
#    affects: A-04 A-05 A-06
#    skips:   A-00 A-01 A-02 A-03

Bugs resolved in the rebuild: cascade delete collection name, LTV division-by-zero, borrowing type typo, payment frequency UI omission, MetricService double-instantiation, expense totals not updating immediately (W2T-44), settings requiring navigate-away-and-back, and more.

Certification condition: One item open — default exchange rate direction (0.74 vs spec 1.42, 1 CAD = X USD vs USD/CAD). All other gaps were low-severity display-order or UX deviations with no data integrity impact.

Artifact Format

All pipeline artifacts are written in .sil format (EMBER — Semantic Intent Language). Plain text. Human readable without a manual. Git diffable. See the EMBER spec for construct definitions.

State File Format

.phoenix/state.json is the pipeline's source of truth. If an agent runs out-of-band (e.g. in Claude Code), use phoenix complete to record it rather than editing the file manually.

{
  "project": "my-project",
  "path": "/path/to/project",
  "createdAt": "2026-04-03T00:00:00.000Z",
  "updatedAt": "2026-04-03T00:00:00.000Z",
  "agents": {
    "a-04": {
      "agentId": "a-04",
      "completedAt": "2026-04-03T12:00:00.000Z",
      "outputCount": 3,
      "confidence": "high",       // "high" | "medium" | "low"
      "summary": "14 signals produced, frequencyFactor bug located"
    }
  },
  "gates": {
    "a-04-approved": {
      "gateId": "a-04-approved",
      "status": "approved",       // "pending" | "approved" | "returned"
      "reviewedAt": "2026-04-03T12:30:00.000Z",
      "notes": "stack recommendation accepted"
    }
  },
  "openEpisodes": ["stack-pivot"]
}

All fields are required. The CLI commands that write state (phoenix complete, phoenix gate) always produce valid records.

phoenix complete a-04 --confidence high --outputs 3 --summary "signals produced"
phoenix validate a-04   # checks artifacts + prints the complete command if clean
phoenix status          # full pipeline view

Build Status

67 tests · 100% statements · 100% functions · 91% branch coverage

Phase 1 — Core          ✓  parser, state, agents, episodes
Phase 2 — Orchestrator  ✓  prompt assembly, artifact validation
Phase 3 — CLI           ✓  init, run, status, gate, episode, validate
Phase 4 — Coverage      ✓  all thresholds met, published to npm

Relationship to cal-runtime

cal-runtime                    phoenix-runtime
───────────────────────────    ────────────────────────────
PEG grammar parser             Line-oriented text parser
Executes CAL scripts           Orchestrates agent pipeline
Produces scores + alerts       Produces prompts + artifact state
npm: @stratiqx/cal-runtime     npm: @semanticintent/phoenix-runtime
bin: cal                       bin: phoenix
Methodology: 6D + DRIFT        Methodology: Phoenix + EMBER

License

MIT