run-talon-smoke - SKILL.md Agent Skill

name: run-talon-smoke description: Run Talon local runtime smoke tests by booting talond with disposable ignored configs, validating Node/better-sqlite3, exercising the terminal channel with talonctl chat, and optionally using the live authenticated codex-cli provider. Use when the user asks for a real-life Talon smoke test, terminal client sanity check, daemon boot/connect verification, local runtime validation, or proof that Talon can receive and answer a terminal message.

Run Talon Smoke

Use this skill to prove that Talon runs locally and the terminal client can communicate with the daemon. Prefer the live codex-cli path when the user asks for a real-life smoke test. Use the fake provider path when the user asks for a no-cost or deterministic smoke test.

Quick Decision

Real-life smoke: use live.
No-cost smoke: use fake.
Broad validation: use both, then still do a manual talonctl chat check for live if the user specifically asked for terminal client proof.

Workflow

Confirm the current branch and working directory.

rtk git status --short --branch

Prepare the selected smoke harness and run doctor.

rtk bash .agents/skills/run-talon-smoke/scripts/run-talon-smoke.sh live --prepare

For the no-cost path, replace live with fake.

Start the selected daemon as a foreground long-running process in its own exec session. Keep that session open while testing.

rtk npm run talond -- --config .codex/talon-smoke/talond-live-codex.yaml --env-file .codex/talon-smoke/empty.env

For fake mode:

rtk npm run talond -- --config .codex/talon-smoke/talond-smoke.yaml --env-file .codex/talon-smoke/empty.env

Connect with the real terminal client in a PTY.

rtk proxy npm run talonctl -- chat --host 127.0.0.1 --port 17701 --token live-terminal-token --client-id codex-live-smoke --persona live

For fake mode:

rtk proxy npm run talonctl -- chat --host 127.0.0.1 --port 17700 --token smoke-terminal-token --client-id codex-smoke --persona smoke

Send the smoke prompt through the interactive terminal session.

Live prompt:

Reply exactly: TALON LIVE OK

Expected live response:

TALON LIVE OK

Fake prompt can be any short message. Expected fake response:

Talon smoke OK: terminal round-trip reached fake codex provider.

Collect evidence before stopping the daemon.

rtk bash .agents/skills/run-talon-smoke/scripts/run-talon-smoke.sh live --status

Use fake --status for the fake path.

Stop the terminal client with Ctrl+C, then stop the foreground daemon session with Ctrl+C. If a smoke daemon is still running, stop it with:

rtk bash .agents/skills/run-talon-smoke/scripts/run-talon-smoke.sh live --stop

Verify no listeners remain.

rtk lsof -nP -iTCP:17700 -sTCP:LISTEN
rtk lsof -nP -iTCP:17701 -sTCP:LISTEN

No output and exit code 1 from lsof means the port is clear.

Helper Script

The helper script recreates ignored smoke files under .codex/talon-smoke if they are missing, creates disposable data directories, runs npm run build, runs talonctl doctor, and retries doctor after npm run rebuild:sqlite when the failure looks like a missing Node 24 better-sqlite3 binding.

Useful commands:

# Prepare files, build, run doctor, and print the foreground daemon/chat commands.
rtk bash .agents/skills/run-talon-smoke/scripts/run-talon-smoke.sh live --prepare

# Run an automated direct WebSocket check. This proves daemon/channel/provider,
# but it does not exercise the talonctl chat client.
rtk bash .agents/skills/run-talon-smoke/scripts/run-talon-smoke.sh live

# Stop any smoke daemon for a mode.
rtk bash .agents/skills/run-talon-smoke/scripts/run-talon-smoke.sh live --stop

Reporting

Report these facts:

branch and working directory
mode used: fake, live, or both
whether npm run rebuild:sqlite was needed
talonctl doctor result
terminal client auth result
exact prompt and response
DB evidence: thread/message/queue/run counts and last outbound body
whether the daemon and chat client were stopped
any remaining gaps, especially if the automated WebSocket fallback was used instead of talonctl chat