/whats-new

name: whats-new
description: Explain what's newly possible — new Claude Code commands after a knowledge-base update, and installed-but-never-used skills. Use when the autopilot announces a knowledge update, or the user asks "what's new", "有什么新功能", "我装了哪些没用过的 skill".

What's newly possible

Two sources, both local. Speak the user's language, lead with benefit, never list for listing's sake.

New commands (after a knowledge-base update)

Read `${CLAUDE_PLUGIN_ROOT}/knowledge/commands.json` and `playbooks.json`. Compare against what the user has actually seen used (counters in `~/.claude/command-autopilot/state.json`). Pick the 2-3 entries most relevant to how this user works (their evidence shows what they do) and present each as: **what it is → what it does for YOU → the one-line way to try it**. Mention the version it arrived in.

Installed-but-unused skills

Read `~/.claude/command-autopilot/skills-index.json`; cross-reference with `state.json` skill counters. Skills with zero invocations are dormant value the user already owns. Pick the 1-2 most relevant to their evidence profile: "You installed X — it can ... — next time you ..., I'll use it automatically."

High-leverage commands the user has never used

The biggest gap for most users is the powerful commands they don't know exist. Read `${CLAUDE_PLUGIN_ROOT}/knowledge/commands.json` + `playbooks.json`, cross-reference `state.json` command counters, and surface the high-leverage ones with zero usage — Workflow orchestration, parallel `/fork`, `/goal`, `/loop`, `/background`. For each: what it is → what it does for THEIR kind of work → the one-line way to try it. This is the "you can't ask for what you don't know exists" gap — lead with it.

Rules

Max 5 items total per run. Relevance over completeness; the rest stays for next time. End with: "Ask again any time — this list updates itself."