[muzimu217]

Why

Upstream issue #50 in vastsa/pi-desktop-plugins asked how pi.session-orchestrator is used — twice. After installing, the plugin shows no panel and no command, and nothing in the shipped SessionTask tool description or README told users (a) that it lives entirely inside Agent chat, or (b) that no special keyword is needed to trigger it. A 0.6.1 release addressing this briefly existed in the marketplace repository before the plugin moved here, but the trigger wording did not survive the 0.7.0 rewrite, so this re-applies it on top of 0.7.0, adapted to the new pull-first semantics.

What changed

• Tool description (manifest.json): leads with explicit when-to-use trigger conditions — parallel work, spawning sub-sessions/workers, delegation, follow-ups — including example phrasings in Chinese and English, plus a negative scope ("Not for ordinary single-session chat or tasks the current session can finish alone"). Everything after the trigger block is the 0.7.0 description, unchanged. This mirrors how official plugin/skill descriptions are written (capability summary + Use when… + examples + negative scope).
• README: adds a bilingual Quick start (how do I use it?) section right after Install — states there is no panel/command/config, gives three example prompts with what actually happens. Adapted to pull-first: "are the workers done?" is answered with result(sessionId, messageId) on the exact delivery, and users are told notifyOnCompletion: true is opt-in.
• Version 0.7.0 → 0.7.1 (description/docs-level change), changelog updated, test version assertion updated.

Verification

node --test test/*.test.mjs   # 38 pass, 0 fail, 0 skipped

The test suite loads main.js with a mocked host and asserts registration uses the exact manifest descriptor, so the shipped schema and the registered tool stay in sync.