Fix for #168: Don't write output on SessionStart for Copilot

[maxfelker]

tl;dr: !isCodex was too broad — Copilot sessions leaked Claude-only content into additionalContext, breaking token validation on Windows.

---

What broke

ponytail-activate.js had two guards that used isCodex to skip Claude-specific behaviour. Both conditions forgot about Copilot — so Copilot sessions fell through into Claude-only paths.

Bug 1 — statusline nudge injected into Copilot context

// before
if (!isCodex) try {

The statusline nudge appends a raw JSON-like fragment to the hook output:

"statusLine": { "type": "command", "command": "powershell -ExecutionPolicy Bypass -File \"...\"" }

Copilot's parser received this inside additionalContext, tried to interpret it as a structured instruction token, and rejected it: "the token is not a valid instructions separator in this version."

The statusline is a Claude Code setting (~/.claude/settings.json). Copilot has no equivalent and should never see this nudge.

Bug 2 — off-mode emitted 'OK' to Copilot

// before
writeHookOutput('SessionStart', 'off', isCodex ? '' : 'OK');

Codex got '' → empty JSON output {}. Copilot got 'OK' → { "additionalContext": "OK" }. In off mode nothing should be injected. Same class of bug, same fix.

What changed

hooks/ponytail-activate.js only — two lines:

// import isCopilot alongside isCodex
const { clearMode, isCodex, isCopilot, setMode, writeHookOutput } = require('./ponytail-runtime');

// Bug 1: scope statusline nudge to Claude only
if (!isCodex && !isCopilot) try {

// Bug 2: off mode emits nothing for both non-Claude hosts
const hookOutput = (isCodex || isCopilot) ? '' : 'OK';
writeHookOutput('SessionStart', 'off', hookOutput);

isCopilot was already exported from ponytail-runtime.js — it just wasn't imported here.

Test

node tests/hooks.test.js — hook compatibility checks passed.

---

on behalf of @maxfelker — ponytail found it, two lines fixed it.

[rajpratham1]

Nice improvement! Adding isCopilot alongside the existing isCodex checks keeps the behavior consistent across supported environments. I also like the introduction of the hookOutput variable—it makes the intent a bit clearer and avoids repeating the conditional. Overall, the changes look clean and make the logic easier to follow. LGTM!