Launches GitHub Copilot CLI when Tuple transcription completes and asks it to summarize the call.

When call-transcription-complete fires, this trigger opens your preferred terminal running copilot inside the call's transcription directory. Copilot reads the whole transcript and writes a tight summary — outcome, decisions, action items, open questions — then stays open for follow-up questions.

What it does

• Reads the full transcript deterministically via Tuple's bundled watcher (tuple-call-watcher.py --catchup), which prints every record across all of the call's session directories (including mid-call restarts).
• Writes a structured summary first, without preamble: Summary, Decisions, Action items, Open questions, Notable context (sections with nothing to say are omitted).
• Answers follow-ups. The prompt is passed via -i (interactive), so after the summary you can keep asking questions and Copilot answers from the transcript.

Permissions

This trigger does not launch Copilot with a blanket auto-approve (--allow-all-tools). It pre-authorizes only the bundled watcher so its one --catchup read runs unattended; everything else falls back to Copilot's normal per-tool approval prompts.

Copilot matches --allow-tool='shell(<stem>:*)' on a command's leading token (the :* suffix means "this stem followed by a space"), so the launcher passes the relative and absolute forms of the watcher path — all pointing at the same script — permitting ./tuple-call-watcher.py --catchup and nothing else. On first launch Copilot asks once whether you trust this directory; accept it to proceed.

Account and org policy

Copilot CLI must be enabled for your account and not restricted by your organization's Copilot policy. If you see Access denied by policy settings, an org/enterprise admin needs to enable the Copilot CLI policy. See your Copilot settings.

Options

Variable	Effect
CALL_SUMMARY_COPILOT_EXTRA_ARGS	Any other copilot flags, shell-quoted (e.g. --effort high, --model gpt-5.3-codex).
PREFERRED_TERM	Force a terminal: ghostty | iterm | alacritty | terminal.
CALL_SUMMARY_COPILOT_DRY_RUN=1	Ship the watcher and write the launcher without opening a terminal (for testing).

Choosing your terminal

By default the trigger opens the first installed of Ghostty → iTerm → Alacritty → Terminal. To force one, set PREFERRED_TERM at the top of call-transcription-complete (or in the environment):

PREFERRED_TERM="iterm"   # ghostty | iterm | alacritty | terminal

The terminal runs launch-call-summary-copilot.command, whose #!/bin/zsh -li shebang sources your ~/.zprofile and ~/.zshrc, so copilot resolves from the same PATH and environment you get in a normal terminal.

Prerequisites

• macOS
• GitHub Copilot CLI installed (npm i -g @github/copilot) so copilot works in a new terminal. Run it once first to sign in (and to clear the one-time directory-trust prompt).
• A Copilot subscription that includes the CLI, not blocked by org policy (see above).
• python3 (the bundled watcher needs it; install with xcode-select --install)
• Tuple transcription enabled for the call

Installation

Drop this directory into your Tuple triggers folder:

~/.tuple/triggers/call-summary-copilot/

The trigger fires the next time call transcription completes.

How it works

When call-transcription-complete fires, Tuple provides TUPLE_TRIGGER_CALL_ARTIFACTS_DIRECTORY, the directory containing the completed call's transcription artifacts. This trigger:

1. Copies the fixed tuple-call-watcher.py and writes call-summary-copilot-prompt.md into that directory.
2. Writes an executable launch-call-summary-copilot.command wrapper into that directory.
3. Opens it in your preferred terminal via open (LaunchServices). No AppleScript, so it triggers no macOS accessibility prompt.
4. The wrapper starts a login-interactive zsh shell, changes to the transcription directory, and runs copilot -i with the prompt and the scoped watcher allowlist (no --allow-all-tools — see Permissions); Copilot reads the transcript and writes the summary.

A PID file (call-summary-copilot.pid) keeps a duplicate summary from launching for the same call.

For local script testing without opening a terminal, set CALL_SUMMARY_COPILOT_DRY_RUN=1. It still ships the watcher and writes the launcher so you can inspect them.