telnyx-phone

Imported from yisding/easycat (src/easycat/cli/scaffold/templates/telnyx-phone/AGENTS.md). Install upstream with npx skills add yisding/easycat --skill telnyx-phone. Copyright stays with the author.

Agent Guide — $PROJECT_NAME

Guidance for coding agents (and humans) working in this scaffolded EasyCat project. The agent is named "$AGENT_NAME"; its entry point is agent.py. Keep agent.py import-safe: make_agent() builds things, and server.py wires it into the call server. Plain, SDK-free helpers belong in tools.py.

Commands

uv sync                                       # Install dependencies
cp .env.example .env                          # Then fill in API keys
uv run easycat doctor --env-file .env         # Preflight checks
uv run easycat doctor --env-file .env --json  # Parseable preflight
uv run --env-file .env uvicorn server:create_app --factory --host 0.0.0.0 --port 8000  # Run
uv run pytest                                 # Offline tests (no API keys needed)
uv run ruff check agent.py server.py tools.py       # Lint
uv run ruff check --fix agent.py server.py tools.py # Auto-fix lint findings
uv run easycat docs                           # Maintained docs routes

Testing and evals

tests/test_agent.py runs offline: tools.py runs for real, agent.py's wiring (name, instructions, tool list) is asserted as built behind pytest.importorskip("agents"), and ScriptedReasoning stands in for the model while EasyCat's real turn machinery (the send_text path, the audio pipeline, journal, and latency metrics) runs end to end. A green run means the app is wired and the pipeline is healthy.

It says nothing about live model quality — an offline stub cannot decide which tool a model calls. Build on it with the helpers in easycat.debug.testing:

• run_text_turn(agent_or_config, "...") drives one real text turn and returns a journal-backed TurnResult.
• run_text_turns(agent_or_config, ["...", "..."]) runs several turns against one session and returns one TurnResult per input.
• run_scripted_audio_turn(agent) drives one turn through the real audio pipeline with scripted stub I/O — it checks pipeline wiring, not speech quality.
• assert_turn_completed / assert_no_error / assert_regex / assert_exact_match / assert_tool_called check journal records — the same helpers work on exported debug bundles via load_bundle.
• assert_latency(result, max_ms=...) budgets turn latency at a percentile (p50/p90/p95/p99).
• assert_llm_judge(result, ...) scores conversational quality with an LLM rubric (needs OPENAI_API_KEY unless you inject judge=).

Conventions

• Python >= 3.11, 4-space indent, 99-char lines (uv run ruff check).
• Keep secrets in .env (gitignored); commit .env.example updates instead so collaborators see every key they need.
• Run uv run easycat doctor --env-file .env before debugging provider issues by hand; run uv run pytest before and after edits.