← the whole session plugin/skills/communicate-what-you-finished/SKILL.md

On task completion, surface the work product to the person you're working with as a list of UX changes — one sentence per change in "when X then a user would see Y because Z but D" form, with emojis flagging which were verified in a running browser/app. Use after finishing any code change, UX fix, or feature work that produced observable user-facing outcomes. Pairs with /update-the-founder when you also need a synthesized status check-in. Triggers: task complete, work done, finished, ship summary, what did you change, list ux changes, completion summary, what got built, hand off.

If you were dispatched from a seed document (a written statement of work an orchestrating agent handed you), use /agent-outcome-summary instead — it produces the structured VERIFIED/UNVERIFIED/BLOCKED format with pasted evidence, which is what a person needs when they weren't watching the session live.

For all other task completions, use this format:

Before your output, print ## WORK_COMPLETE on its own line, if emit_work_complete_marker is enabled in your alignment-harness config (alignment-harness config path — on by default, costs nothing). This marker lets anything that collects completion reports — a local script, a dashboard, whatever the person builds — find every completion in session transcripts (grep -l "## WORK_COMPLETE" ~/.claude/projects/*/*.jsonl) without parsing them.

On task completion:

List all UX changes you made.

when x then a user would see y because z but d (d = the catch — a limitation or condition the change still depends on)

one sentence each

use an emoji to show which of these you actually checked in the running product — a browser, a simulator, a CLI, whatever surface a real user touches — to confirm it works as intended. Be specific.

If you have no way to check the running product (no browser tool connected, no local server, a library with no UI): still write the one-sentence list, but mark every item as not yet checked and say so in words. Never drop the emoji silently, and never mark something checked when you only read the code.