| name | phoenix-setup |
| description | Install the ATV-Phoenix self-healing harness into GitHub Copilot. Builds the Rust MCP server (if needed) and registers the phoenix agent plus its MCP server so Copilot can call phoenix_sense / phoenix_snapshot / phoenix_heal / phoenix_verify_trace. Use when the user types /phoenix-setup or asks to install or enable Phoenix. |
/phoenix-setup
This skill installs ATV-Phoenix into the GitHub Copilot CLI so the phoenix agent and its four
self-healing tools become available.
What it does
- Ensures the Rust binary
phoenix-mcp exists (builds it with cargo build --release if missing).
- Registers the
phoenix MCP server in ~/.copilot/mcp-config.json (so the tools auto-load).
- Installs the
phoenix agent definition to ~/.copilot/agents/phoenix.agent.md (so
copilot --agent phoenix works after a restart).
- Installs the 13 bundled skills into
~/.copilot/skills, then self-checks them with
phoenix-mcp doctor (the harness validates its own skills).
- Installs the bundled TokenMasterX for token-efficient retrieval (best-effort; needs
graphify).
How to run it
The setup script ships in this skill's directory. Locate it, then run it with the repo root:
- Run
/skills info phoenix-setup and read the skill's directory path from the output.
- Run the setup script, passing the skill dir and the repo root:
python <skill-dir>/setup.py --repo <ATV-Phoenix-repo-root>
If --repo is omitted, the script looks for the repo via the PHOENIX_REPO env var or the current
directory. It prints exactly what it registered and the restart command.
After install
Restart Copilot (or start with copilot --agent phoenix). Then ask Copilot to verify and self-heal a
task — it will call phoenix_sense / phoenix_heal and only report success on objective evidence.
Honesty
The tools report objective signals only (exit codes / hashes / regex). A fabricated "done" is the
failure mode Phoenix prevents; "I'm not sure it passed" is acceptable.
