Topping Out: Status refresher

Give the owner the big picture in fifteen seconds of reading. For the moment they step back, or step away and come back lost.

Usage

/f-status              # refresh on the whole session
/f-status <topic>      # narrow to one thread

Hard constraint: conversation only

Read only what is already in this session's context. This skill must answer instantly.

Forbidden here: subagents, git commands, file reads, file writes, MCP calls, web fetches, any tool at all. If you reach for a tool, the skill has failed, the owner wanted a refresher, not an investigation.

You already know what happened. Say it.

If the session was compacted, work from the summary and say so in one clause (earlier detail is summarised). Never invent specifics to fill a gap.

If the session is genuinely thin, just started, nothing done yet, say that in two lines and stop. A short honest answer beats a padded one.

Output

Five blocks, in this order. Skip any block with nothing real in it. Hard cap ~200 words.

Where we are
<one sentence a non-engineer understands>

Done this session
 • <what changed in the world>
 • <3-5 bullets max>

Right now
<health line>

Next
<the single next action, and who does it, you or me. If nothing is needed from them: `nothing from you`>

Watch out
<only a real trap; otherwise omit entirely>

Health line uses the house vocabulary: Healthy, <scope>: <evidence> · Blocked, <scope>: <what's in the way> · Waiting, <scope>: <on what> · Unknown, <scope>: <what we'd need to check>. Name the scope you actually checked. Never imply whole-system health from one green gate.

Language rules

Write it the way you'd explain it to a sharp friend who missed the last hour.

• Plain words. No jargon unless the owner has been using that word themselves this session. If a term is load-bearing and new to them, define it in four words inline.
• One idea per bullet. If a bullet needs a comma-spliced second clause, it's two bullets or it's cut.
• What changed in the world, not what tool ran. the login page now remembers you, not edited auth-core.ts and ran typecheck.
• No file paths, no command names, no SHAs in the body. They raise cognitive load and the owner can ask.
• Numbers beat adjectives where a number exists. 4 of 7 tests pass, not mostly passing.
• No process replay. They don't need the order things happened in, only where it landed.
• No reasoning. State the situation; don't re-derive it.
• Status is not an action. Right now is state. Next is a thing the owner does. Never write a non-action in imperative form (keep X, leave Y alone, don't touch Z) inside Next, keep X is shaped exactly like delete X, so it reads as a task when it is the absence of one. If there is nothing for them to do, Next reads nothing from you and stops.
• Say a caveat once. A constraint already stated earlier in the session is a settled fact, not an open item. Repeating it makes it look unmet.

What this is not

Not a report card. Don't congratulate, don't hedge, don't apologise for what isn't done. If something failed, one flat line: what failed, and whether it blocks the next step.

Don't re-open settled decisions. If the owner already chose something, it's a fact now, not an option.

Don't append offers, next-step menus, or "let me know if…". End on Next.

Example

Where we are We're building a small skill that tells you where you are, so you can step away and come back without re-reading everything.

Done this session • Checked two public skill collections, neither had what you wanted, but both had usable parts • Settled the design: plain English, five blocks, reads only this conversation • Moved off the main branch so nothing can reach production by accident

Right now Healthy, the new skill file: written and mirrored, nothing committed yet.

Next You read it and tell me if the tone is right. Nothing is staged until you say so.