MarkRoyal44

When To Skill — AI skill for Claude Code

AI community

A skill that knows when to run AI skills for you.

How to install When To Skill

This entry records only its repository, not the path inside it, so there is no exact command to give. Open MarkRoyal44/when-to-skill and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What When To Skill does

A skill that knows when to run AI skills for you. Use skill at beginning of session, It'll lay dormant and eventually tells you which skills should have fired, which you were missing, which ran and changed nothing, and which pair better chained. Plain markdown — works with Claude Code, Claude.ai, Cursor, or anything squishy that reads skill files.

Alternatives in AI

CLI Continues — resume any AI coding session in another tool — Claude Code, Copilot, Gemini, Codex, Cursor 1.5k ★
Firm — Convene your standing AI staff — memos on every beat, a board session without you, minutes with dissent preser 1.3k ★
Botmux — Bridge Feishu/Lark to AI coding CLIs — Claude Code, Codex, Gemini, OpenCode… every DM, group or topic spawns i 1.2k ★

README

👁️ when-to-skill

**A dormant auditor for your AI skills. It costs nothing until it has something real to say.**

You install a pile of skills. Six months later, do you know which ones actually fire? Which ones you needed and never had? Which one has been quietly running and changing nothing?

Nobody knows. **This is the thing that knows.**

> arm the auditor
👁️ Armed. Dormant until the sample is real.

Then you forget about it — **being forgotten is the success condition.** A watcher that announces itself changes the behaviour it was armed to observe.

What it eventually tells you

👁️ SKILL AUDIT — 7 scans, 3 tasks

✅ Should have used:
- the "make it modular" refactor → `grill-me` — would have caught that
  "modular" meant file split, not dependency inversion

🕳️ Missing entirely:
- verifying that file writes actually landed — seen 3x, nothing covers it → candidate

🗑️ Ran and didn't earn it:
- `changelog-writer` — fired 5x, changed nothing you kept

🔗 Stronger together:
- `grill-me` → `gap-radar` — grill closes the gap before the build,
  radar catches what survived. Seen adjacent 2x, never chained.

🌱 Worth building:
- "did the thing I just built actually work" — seen 4x, always done by hand

Why the 🗑️ section is the important one

**A tool that only ever recommends more of itself is a salesman.**

Naming a skill that wasted your tokens is what makes the other four sections believable — and it is the only mechanism in your setup that can make your skill collection *smaller* instead of larger.

The design rule

⚡ **It is never allowed to cost a round trip.**

No turn of its own. It appends one paragraph to a reply you were already getting, or it says nothing.
No timer, no daemon, no scheduled check-in.
Zero findings = it never wakes at all. Silence is a valid outcome, not a failure.
Ask once. Declined is final for the session. Nagging is how a good tool gets uninstalled.

The only tokens it