Project Phoenix

Appearance

Phoenix Runtime ​

@semanticintent/phoenix-runtime is the open-source CLI that orchestrates the seven-agent pipeline. It manages .sil artifact state, enforces human gates, injects episode context, and produces agent prompts ready to run in any AI interface.

DOI

Install ​

bash
npm install @semanticintent/phoenix-runtime

How It Works ​

The runtime does not call an AI API. It produces prompts. You paste them into Claude Code, Claude CLI, or any AI interface. Model-agnostic. No API key management.

phoenix run a-01
  → checks state.json (prerequisites met?)
  → reads open episodes (any affect a-01?)
  → loads A-01-BUSINESS-LOGIC-EXTRACTOR.md
  → assembles: episode context + project context + agent instructions
  → outputs: ready-to-run prompt

State lives in .phoenix/state.json — plain JSON, human readable, git diffable, travels with the project.

CLI Reference ​

bash
# Initialise a new project
phoenix init <project-name>

# Run an agent — produces the prompt
phoenix run <agent-id>
phoenix run a-01 --output prompt.md

# Show pipeline status
phoenix status

# Approve or return a human gate
phoenix gate <gate-id> --approve --notes "looks correct"
phoenix gate <gate-id> --return  --notes "missing error state"

# Episode management
phoenix episode new             # record a mid-engagement change
phoenix episode list
phoenix episode resolve <id>

# Validate artifacts for an agent
phoenix validate <agent-id>

EMBER Artifact Format ​

All pipeline artifacts are written in .sil format (EMBER — Semantic Intent Language). Five construct types: signal, workflow, screen, spec, episode.

Full language specification: EMBER — Semantic Intent Language

Citation ​

If you use the runtime in your work:

bibtex
@software{shatny2026phoenixruntime,
  author = {Shatny, Michael},
  title  = {Phoenix Pipeline Runtime},
  year   = {2026},
  doi    = {10.5281/zenodo.19360782},
  url    = {https://github.com/semanticintent/phoenix-runtime}
}